OpenMMLab Computer Vision Foundation

Last update: Jan 09, 2023

Related tags

Deep Learning mmcv

Overview

English | 简体中文

Introduction

MMCV is a foundational library for computer vision research and supports many research projects as below:

MMCV: OpenMMLab foundational library for computer vision.
MIM: MIM Installs OpenMMLab Packages.
MMClassification: OpenMMLab image classification toolbox and benchmark.
MMDetection: OpenMMLab detection toolbox and benchmark.
MMDetection3D: OpenMMLab's next-generation platform for general 3D object detection.
MMSegmentation: OpenMMLab semantic segmentation toolbox and benchmark.
MMAction2: OpenMMLab's next-generation action understanding toolbox and benchmark.
MMTracking: OpenMMLab video perception toolbox and benchmark.
MMPose: OpenMMLab pose estimation toolbox and benchmark.
MMEditing: OpenMMLab image and video editing toolbox.
MMOCR: A Comprehensive Toolbox for Text Detection, Recognition and Understanding.
MMGeneration: OpenMMLab image and video generative models toolbox.
MMFlow: OpenMMLab optical flow toolbox and benchmark.

It provides the following functionalities.

Universal IO APIs
Image/Video processing
Image and annotation visualization
Useful utilities (progress bar, timer, ...)
PyTorch runner with hooking mechanism
Various CNN architectures
High-quality implementation of common CUDA ops

See the documentation for more features and usage.

Note: MMCV requires Python 3.6+.

Installation

There are two versions of MMCV:

mmcv-full: comprehensive, with full features and various CUDA ops out of box. It takes longer time to build.
mmcv: lite, without CUDA ops but all other features, similar to mmcv<1.0.0. It is useful when you do not need those CUDA ops.

Note: Do not install both versions in the same environment, otherwise you may encounter errors like ModuleNotFound. You need to uninstall one before installing the other. Installing the full version is highly recommended if CUDA is available.

a. Install the full version.

Before installing mmcv-full, make sure that PyTorch has been successfully installed following the official guide.

We provide pre-built mmcv packages (recommended) with different PyTorch and CUDA versions to simplify the building. In addition, you can run check_installation.py to check the installation of mmcv-full after running the installation commands.

i. Install the latest version.

The rule for installing the latest mmcv-full is as follows:

pip install mmcv-full -f https://download.openmmlab.com/mmcv/dist/{cu_version}/{torch_version}/index.html

Please replace {cu_version} and {torch_version} in the url to your desired one. For example, to install the latest mmcv-full with CUDA 11.1 and PyTorch 1.9.0, use the following command:

pip install mmcv-full -f https://download.openmmlab.com/mmcv/dist/cu111/torch1.9.0/index.html

Note: mmcv-full is only compiled on PyTorch 1.x.0 because the compatibility usually holds between 1.x.0 and 1.x.1. If your PyTorch version is 1.x.1, you can install mmcv-full compiled with PyTorch 1.x.0 and it usually works well. For example, if your PyTorch version is 1.8.1 and CUDA version is 11.1, you can use the following command to install mmcv-full.

pip install mmcv-full -f https://download.openmmlab.com/mmcv/dist/cu111/torch1.8.0/index.html

For more details, please refer the the following tables and delete =={mmcv_version}.

ii. Install a specified version.

The rule for installing a specified mmcv-full is as follows:

pip install mmcv-full=={mmcv_version} -f https://download.openmmlab.com/mmcv/dist/{cu_version}/{torch_version}/index.html

First of all, please refer to the Releases and replace {mmcv_version} a specified one. e.g. 1.3.9. Then replace {cu_version} and {torch_version} in the url to your desired versions. For example, to install mmcv-full==1.3.9 with CUDA 11.1 and PyTorch 1.9.0, use the following command:

pip install mmcv-full==1.3.9 -f https://download.openmmlab.com/mmcv/dist/cu111/torch1.9.0/index.html

For more details, please refer the the following tables.

CUDA	torch1.10	torch1.9	torch1.8	torch1.7	torch1.6	torch1.5
11.3	install `pip install mmcv-full=={mmcv_version} -f https://download.openmmlab.com/mmcv/dist/cu113/torch1.10.0/index.html`
11.1	install `pip install mmcv-full=={mmcv_version} -f https://download.openmmlab.com/mmcv/dist/cu111/torch1.10.0/index.html`	install `pip install mmcv-full=={mmcv_version} -f https://download.openmmlab.com/mmcv/dist/cu111/torch1.9.0/index.html`	install `pip install mmcv-full=={mmcv_version} -f https://download.openmmlab.com/mmcv/dist/cu111/torch1.8.0/index.html`
11.0				install `pip install mmcv-full=={mmcv_version} -f https://download.openmmlab.com/mmcv/dist/cu110/torch1.7.0/index.html`
10.2	install `pip install mmcv-full=={mmcv_version} -f https://download.openmmlab.com/mmcv/dist/cu102/torch1.10.0/index.html`	install `pip install mmcv-full=={mmcv_version} -f https://download.openmmlab.com/mmcv/dist/cu102/torch1.9.0/index.html`	install `pip install mmcv-full=={mmcv_version} -f https://download.openmmlab.com/mmcv/dist/cu102/torch1.8.0/index.html`	install `pip install mmcv-full=={mmcv_version} -f https://download.openmmlab.com/mmcv/dist/cu102/torch1.7.0/index.html`	install `pip install mmcv-full=={mmcv_version} -f https://download.openmmlab.com/mmcv/dist/cu102/torch1.6.0/index.html`	install `pip install mmcv-full=={mmcv_version} -f https://download.openmmlab.com/mmcv/dist/cu102/torch1.5.0/index.html`
10.1			install `pip install mmcv-full=={mmcv_version} -f https://download.openmmlab.com/mmcv/dist/cu101/torch1.8.0/index.html`	install `pip install mmcv-full=={mmcv_version} -f https://download.openmmlab.com/mmcv/dist/cu101/torch1.7.0/index.html`	install `pip install mmcv-full=={mmcv_version} -f https://download.openmmlab.com/mmcv/dist/cu101/torch1.6.0/index.html`	install `pip install mmcv-full=={mmcv_version} -f https://download.openmmlab.com/mmcv/dist/cu101/torch1.5.0/index.html`
9.2				install `pip install mmcv-full=={mmcv_version} -f https://download.openmmlab.com/mmcv/dist/cu92/torch1.7.0/index.html`	install `pip install mmcv-full=={mmcv_version} -f https://download.openmmlab.com/mmcv/dist/cu92/torch1.6.0/index.html`	install `pip install mmcv-full=={mmcv_version} -f https://download.openmmlab.com/mmcv/dist/cu92/torch1.5.0/index.html`
cpu	install `pip install mmcv-full=={mmcv_version} -f https://download.openmmlab.com/mmcv/dist/cpu/torch1.10.0/index.html`	install `pip install mmcv-full=={mmcv_version} -f https://download.openmmlab.com/mmcv/dist/cpu/torch1.9.0/index.html`	install `pip install mmcv-full=={mmcv_version} -f https://download.openmmlab.com/mmcv/dist/cpu/torch1.8.0/index.html`	install `pip install mmcv-full=={mmcv_version} -f https://download.openmmlab.com/mmcv/dist/cpu/torch1.7.0/index.html`	install `pip install mmcv-full=={mmcv_version} -f https://download.openmmlab.com/mmcv/dist/cpu/torch1.6.0/index.html`	install `pip install mmcv-full=={mmcv_version} -f https://download.openmmlab.com/mmcv/dist/cpu/torch1.5.0/index.html`

Note: The pre-built packages provided above do not include all versions of mmcv-full, you can click on the corresponding links to see the supported versions. For example, you can click cu102-torch1.8.0 and you can see that cu102-torch1.8.0 only provides 1.3.0 and above versions of mmcv-full. In addition, We no longer provide mmcv-full pre-built packages compiled with PyTorch 1.3 & 1.4 since v1.3.17. You can find previous versions that compiled with PyTorch 1.3 & 1.4 here. The compatibility is still ensured in our CI, but we will discard the support of PyTorch 1.3 & 1.4 next year.

Another way is to compile locally by running

pip install mmcv-full

Note that the local compiling may take up to 10 mins.

b. Install the lite version.

pip install mmcv

c. Install full version with custom operators for onnxruntime

Check here for detailed instruction.

If you would like to build MMCV from source, please refer to the guide.

FAQ

If you face some installation issues, CUDA related issues or RuntimeErrors, you may first refer to this Frequently Asked Questions.

Citation

If you find this project useful in your research, please consider cite:

@misc{mmcv,
    title={{MMCV: OpenMMLab} Computer Vision Foundation},
    author={MMCV Contributors},
    howpublished = {\url{https://github.com/open-mmlab/mmcv}},
    year={2018}
}

Contributing

We appreciate all contributions to improve MMCV. Please refer to CONTRIBUTING.md for the contributing guideline.

License

MMCV is released under the Apache 2.0 license, while some specific operations in this library are with other licenses. Please refer to LICENSES.md for the careful check, if you are using our code for commercial matters.

Comments

Support receptive field search of CNN models. (TPAMI paper rf-next)
Motivation

Merging a general receptive field search method to mmcv and mmdet. (paper: RF-Next: Efficient Receptive Field Search for Convolutional Neural Networks TPAMI 2022 pdf)

Modification

The RFSearch module is included.

BC-breaking (Optional)

Does the modification introduce changes that break the backward-compatibility of the downstream repositories? If so, please describe how it breaks the compatibility and how the downstream projects should modify their code to keep compatibility with this PR.

Use cases (Optional)

If this PR introduces a new feature, it is better to list some use cases here, and update the documentation.

This pr is a part of pr in mmdet https://github.com/open-mmlab/mmdetection/pull/8191

Checklist

Before PR:

[x] I have read and followed the workflow indicated in the CONTRIBUTING.md to create this PR.

[x] Pre-commit or linting tools indicated in CONTRIBUTING.md are used to fix the potential lint issues.

[ ] Bug fixes are covered by unit tests, the case that causes the bug should be added in the unit tests.

[ ] New functionalities are covered by complete unit tests. If not, please add more unit test to ensure the correctness.

[ ] The documentation has been modified accordingly, including docstring or example tutorials.

After PR:

[x] If the modification has potential influence on downstream or other related projects, this PR should be tested with some of those projects, like MMDet or MMCls.

[x] CLA has been signed and all committers have signed the CLA in this PR.
opened by gasvn 36
Easy contributions! Add type annotation for mmcv~
Hi guys, we are planning to add type annotation for python code in MMCV. PRs of adding type annotation from community are welcomed :stuck_out_tongue_closed_eyes:. To make the community fully aware of the progress, we list the progress here. Please feel free to leave a message and create a PR if you are willing to add type annotation for one of the following modules :rocket::rocket:.

[x] runner part1

[x] base_runner.py #2003

[x] epoch_based_runner.py #2003

[x] iter_based_runner.py #2003

[x] base_module.py #2020

[x] builder.py @ytzhao #1985

[x] checkpoint.py #2020

[x] default_constructor.py #2020

[x] dist_utils.py #2020

[x] fp16_utils.py #2020

[x] log_buffer.py #2020

[x] priority.py #2020

[x] utils.py #2008

[x] optimizer #2008

[x] runner part2

[x] checkpoint.py @triwahyuu #1991

[x] closure.py @triwahyuu #1991

[x] ema.py @triwahyuu #1991

[x] evaluation.py @triwahyuu #1991

[x] hook.py @triwahyuu #1991

[x] iter_timer.py @triwahyuu #1991

[x] lr_updater.py @WINDSKY45 #1988

[x] memory.py @triwahyuu #1991

[x] momentum_updater.py @triwahyuu #1991

[x] optimizer.py @triwahyuu #1991

[x] sync_buffer.py @triwahyuu #1991

[x] runner part3

[x] logger #2000

[x] ops part1

[x] active_rotated_filter.py #2017

[x] assign_score_withk.py #2023

[x] ball_query.py@ShawnHXH #2037 #1981

[x] bbox.py @RangeKing https://github.com/open-mmlab/mmcv/pull/1981

[x] border_align.py@ShawnHXH #2037

[x] box_iou_rotated.py @RangeKing https://github.com/open-mmlab/mmcv/pull/1981

[x] carafe.py #2032

[x] cc_attention.py @zengxiang68 #1987

[x] contour_expand.py @zengxiang68 #1987

[x] convex_iou.py @zengxiang68 #1987

[x] corner_pool.py #2032

[x] correlation.py@ShawnHXH #2037

[x] deform_conv.py@ShawnHXH #2037

[x] deform_roi_pool.py@ShawnHXH #2037

[x] deprecated_wrappers.py@ShawnHXH #2037

[x] diff_iou_rotated.py #2032

[x] ops part2

[x] focal_loss.py @WINDSKY45 #1994

[x] furthest_point_sample.py @WINDSKY45 #1994

[x] fused_bias_leakyrelu.py #2030

[x] gather_points.py#2030

[x] group_points.py#2030

[x] info.py@WINDSKY45 #2030

[x] iou3d.py @gy-7 #1989

[x] knn.py@WINDSKY45 #2030

[x] masked_conv.py@WINDSKY45 #2030

[x] merge_cells.py@WINDSKY45 #2030

[x] min_area_polygons.py@WINDSKY45 #2030

[x] modulated_deform_conv.py@WINDSKY45 #2030

[x] multi_scale_deform_attn.py @WINDSKY45 #2030

[x] nms.py @liuyanyi #2006

[x] pixel_group.py #2016

[x] point_sample.py #2019

[x] points_in_boxes.py #2007

[x] points_in_polygons.py #2013

[x] points_sampler.py #2015

[x] ops part3

[x] psa_mask.py #1995

[x] riroi_align_rotated.py #1995

[x] roi_align.py #1995

[x] roi_align_rotated.py #1995

[x] roi_pool.py #1995

[x] roiaware_pool3d.py #1995

[x] roipoint_pool3d.py #1995

[x] rotated_feature_align.py #1995

[x] saconv.py #1995

[x] scatter_points.py #1995

[x] sparse_conv.py #1995

[x] sparse_functional.py #1995

[x] sparse_modules.py #1995

[x] sparse_ops.py #1995

[x] sparse_pool.py #1995

[x] sparse_structure.py #1995

[x] sync_bn.py #1995

[x] three_interpolate.py #1995

[x] three_nn.py #1995

[x] tin_shift.py #1995

[x] upfirdn2d.py #1995

[x] voxelize.py #1995

[x] bricks #1993

[x] cnn/utils #1996

[ ] net

[x] alexnet.py @yangrisheng #1983

[ ] builder.py @triple-Mu #1980

[x] resnet.py #1982

[x] vgg.py @nxznm #2050

[x] fileio

[x] handlers #1997

[x] file_client #1942

[x] io #1997

[x] parse #1997

[ ] utils

[x] parallel #2031

[x] engine #2034

[x] video

[x] io #1950

[x] optflow #1950

[x] processing #1999

[x] visualization

[x] image.py #1946

[x] optflow.py #1947

[x] color.py #1947

[ ] image

[x] colorspace.py #1999

[ ] geometric.py #2063

[ ] io.py #2063

[ ] misc.py #2063

[x] photometric.py #1999

[x] tensorrt #1950

[x] onnx #1999

[x] quantization #1950

[ ] device #1990
opened by HAOCHENYE 33
Installation takes forever
Hi guys,

I am running

pip install mmcv-full -f https://download.openmmlab.com/mmcv/dist/cu110/torch1.9.0/index.html

On a quite powerful 8 core machine and the installation takes forever, more than 30m. The process is stuck at Building wheel for mmcv-full (setup.py) ... |

I am wondering if this is okay.

Thank you :)

Francesco
Installation
opened by FrancescoSaverioZuppichini 29
[Feature] Add CRF module
Thanks for your contribution and we appreciate it a lot. The following instructions would make your pull request more healthy and more easily get feedback. If you do not understand some items, don't worry, just make the pull request and seek help from maintainers.

Motivation

add the crf block, and crf can improve the model capability

Modification

add the crf module and test_crf.py

BC-breaking (Optional)

Does the modification introduce changes that break the backward-compatibility of the downstream repositories? If so, please describe how it breaks the compatibility and how the downstream projects should modify their code to keep compatibility with this PR.

Use cases (Optional)

If this PR introduces a new feature, it is better to list some use cases here, and update the documentation.

Checklist

Before PR:

[x] I have read and followed the workflow indicated in the CONTRIBUTING.md to create this PR.

[ ] Pre-commit or linting tools indicated in CONTRIBUTING.md are used to fix the potential lint issues.

[ ] Bug fixes are covered by unit tests, the case that causes the bug should be added in the unit tests.

[ ] New functionalities are covered by complete unit tests. If not, please add more unit test to ensure the correctness.

[ ] The documentation has been modified accordingly, including docstring or example tutorials.

After PR:

[x] If the modification has potential influence on downstream or other related projects, this PR should be tested with some of those projects, like MMDet or MMCls.

[ ] CLA has been signed and all committers have signed the CLA in this PR.

Waiting for Design
opened by 459737087 28
runner constructor

Add RunnerConstructor feature to build_runner.

It will not effect the original code feature, but just an enhancement.

Refer here for more details: issue1225

opened by densechen 25
ModuleNotFoundError: No module named 'mmcv._ext'

I clone the repo and run pip install.

When I run test_image.py, this error shows up.

https://github.com/open-mmlab/mmcv/blob/120c6a64daefe97d451cbea4eae15af0ade50a17/mmcv/video/optflow.py#L4

opened by Liyw979 22
[Fix] Restrict the warning message
Thanks for your contribution and we appreciate it a lot. The following instructions would make your pull request more healthy and more easily get feedback. If you do not understand some items, don't worry, just make the pull request and seek help from maintainers.

Motivation

When I implemented a CenterNet2 model in mmdet, there was a situation when I had to use GroupNorm after Conv2d, and the Conv2d layer was supposed to have a bias. Once I finished the modeling part, I got a warning message 'ConvModule has norm and bias at the same time'. At first, I was not so sure about this warning, it made me uneased. Then I did a little research and realized that for BatchNorm (not so sure about SyncBN), the bias does not affect the calculation. But for other kinds of Norm, e.g., GN, I am pretty sure the bias does give some influence. I think maybe we should restrict the popping of this warning message, since it can be quite misleading for new starters.

Modification

Restrict a warning message of 'ConvModule has norm and bias at the same time' to BN, BN1, BN2 and BN3 normalization.

BC-breaking (Optional)

Does the modification introduce changes that break the backward-compatibility of the downstream repos? If so, please describe how it breaks the compatibility and how the downstream projects should modify their code to keep compatibility with this PR.

Use cases (Optional)

If this PR introduces a new feature, it is better to list some use cases here, and update the documentation.

Checklist

Pre-commit or other linting tools are used to fix the potential lint issues.

The modification is covered by complete unit tests. If not, please add more unit test to ensure the correctness.

If the modification has potential influence on downstream projects, this PR should be tested with downstream projects, like MMDet or MMCls.

The documentation has been modified accordingly, like docstring or example tutorials.
opened by yyz561 19
[Feature] Add EvalHook which will be used in downstream projects
Overview

This PR adds mmcv/runner/hooks/eval.py file and EvalHook and DistEvalHook to MMCV together with related unittest parts. Since those two hooks have been widely used in many MM repos, this PR implements and unifies common functions used in the evaluation part.

Design

Model Evaluation performed in OpenMMLab uses EvalHook and DistEvalHook during training.

In detail, the those hooks are commonly registered in apis/train.py by runner.register_hook(eval_hook(val_dataloader, **eval_cfg)). Users can initialize their models with following steps:

Define validation dataset by assigning cfg.data.val, building val_datasetloader

Register EvalHook using runner.register_hook.

Once registering it, runner will periodically call the evaluate function to perform evaluation in a fixed mode (by epoch or by iteration, etc.). The high-level workflow of EvalHook in OpenMMLab is:

register EvalHook -> define evaluation setting (start, interval, by_epoch, save_best, comparison rules, etc.) -> perform evaluation after_train_iter or after_train_epoch -> (save the best checkpoint) -> loop back to after_train_xxx

APIs

EvalHook is the base module, and DistEvalHook is a child class to it. They both inherit Hook. Here is the APIs explanation for EvalHook.

Initialization

dataloader: The PyTorch dataloader for evaluation. It can be built with build_dataloader with provided dataloader setting.

start: Evaluation starting epoch. It enables evaluation before the training starts if start <= the resuming epoch. If set to None, whether to evaluate is merely decided by interval.

interval: Evaluation interval.

by_epoch: Determine whether to perform evaluation by epoch or by iteration. If set to True, it will perform by epoch. Otherwise, by iteration.

save_best: If a metric is specified, it would measure the best checkpoint during evaluation. The information about best checkpoint would be save in runner.meta['hook_msgs']. Options are the evaluation metrics to the validation dataset. e.g., bbox_mAP, segm_mAP for bbox detection and instance segmentation. [email protected] for proposal recall. If save_best is auto, the first key of the returned OrderedDict result will be used. The interval of CheckpointHook should be divisible by that of EvalHook.

rule: Comparison rule for best score. If set to None, it will infer a reasonable rule. Keys such as acc, top, etc. will be inferred by greater rule. Keys contain loss will be inferred by less rule. Options are greater, less, None.

But Note: Since the rule for downstream repos are different, it may need to overwrite the self.greater_keys and self.less_keys

eval_kwargs: Key word arguments for dataset evaluate function (def evaluate), which will be fed into the evaluate function of the dataset.

Other hardcoded inner variables: rule_map, init_value_map, greater_keys and less_keys.

Note that: Since the rule for downstream repos are different, it may need to overwrite these variable due to the specific task.

rule_map: A dict containing comparison function, default as {'greater': lambda x, y: x > y, 'less': lambda x, y: x < y}.

init_value_map: The initialized value for comparison, default as {'greater': -inf, 'less': inf}.

greater_keys: A list containing some rule keys applied for greater function, which means in these rules EvalHook regards greater number as the better one. Note that: If a string is one of the rules' substring, it will be applied to greater function. e.g., for acc: top1_acc, top5_acc, mean_acc are all applied to greater rule.

less_keys: Similar to greater_keys but it is for less rules. e.g., for loss: bce_loss, bmn_loss are all applied to less rule.

before_run

This part is to initialize runner.meta.hook_msgs, if users determine to save the best checkpoint.

before_train_xxx

For before_train_iter and before_train_epoch part. It is mainly to determine whether it is the right time to perform evaluation by examining:

Whether it is the correct time to perform evaluation according to by_epoch

Whether it achieves the start to perform evaluation

And use self.initial_flag to indicate whether the EvalHook gets into the normal evaluation loop. After getting into the normal evaluation loop, before_train_xxx will be skipped.

after_train_xxx

For after_train_iter and after_train_epoch part. It is mainly to inference the model and do evaluation, as well as save the best checkpoint if save_best is specified. In detail, it will call _do_evaluate() and _save_best().

_do_evaluate(): inference the model by calling single_gpu_test(), do evaluation and call _save_best()

_save_best(): compare the score according to the rule, write info into runner.meta['hook_msgs'] and save the best checkpoint in the work_dir

For DistEvalHook, besides the variable and function mentioned above, it inferences model by calling multi_gpu_test() and assigns tmpdir, gpu_collect for multi_gpu_test()

Usages

Users adopt EvalHook by typing in --validate for training command, it will call EvalHook in training.

Config

To use EvalHook with a specific setting, users need to modify the evaluation variable in config files. like this:

evaluation = dict( interval=1, metrics=['top_k_accuracy', 'mean_class_accuracy'])

the key-value pairs in the dict are corresponding to EvalHook API.

Migration

Since the key for determine greater an less is related to the downstream task, downstream repos may need to overwrite the self.greater_keys and self.less_keys:

from mmcv.runner import EvalHook as BasicEvalHook class EvalHook(BasicEvalHook): greater_keys = ['a', 'b'] less_keys = ['c'] def __init__(self, *args, **kwargs): super().__init__(*args, **kwargs)
opened by dreamerlin 19
TypeError: init_weights() missing 1 required positional argument: 'pretrained'

File "tools/train.py", line 189, in main() File "tools/train.py", line 163, in main model.init_weights() File "C:\Users\topseven\anaconda3\envs\mmcv\lib\site-packages\mmcv\runner\base_module.py", line 117, in init_weights m.init_weights() TypeError: init_weights() missing 1 required positional argument: 'pretrained'
Downstream

opened by apanand14 17

Initialization makes no effect.

I attempt to modify the initialization of the weight of MultiScaleDeformableAttention and implement it as follows. I assign constant 0 values to weight and bias of sampling_offset layer. While initialization, the program goes into reset_parameters function and print 0 as expected. However, at the first forward step, the program prints a large value. I suspect anothre initialization is invoked somewhere. How to address this issue?

import torch
import torch.nn as nn

from mmdet.core.bbox import bbox_cxcywh_to_xyxy
from mmcv.cnn import constant_init, xavier_init
from mmcv.cnn.bricks.registry import ATTENTION
from mmcv.ops.multi_scale_deform_attn import MultiScaleDeformableAttnFunction, multi_scale_deformable_attn_pytorch
from mmcv.runner.base_module import BaseModule
from mmcv.cnn.bricks.transformer import MultiScaleDeformableAttention


@ATTENTION.register_module()
class MultiScaleRectifiedAttention(MultiScaleDeformableAttention):
    def __init__(self,
                 embed_dims=256,
                 num_heads=8,
                 num_levels=4,
                 num_points=4,
                 im2col_step=64,
                 dropout=0.1,
                 norm_cfg=None,
                 init_cfg=None):
        super(MultiScaleRectifiedAttention, self).__init__(embed_dims, num_heads, num_levels,
                                                           num_points, im2col_step, dropout,
                                                           norm_cfg, init_cfg)

        self.reset_parameters()

    def reset_parameters(self):
        constant_init(self.sampling_offsets, val=0., bias=0.)
        constant_init(self.attention_weights, val=0., bias=0.)

        print(self.sampling_offsets.weight.abs().sum())
        print(self.sampling_offsets.bias.abs().sum())

    def forward(self,
                query,
                key,
                value,
                residual=None,
                query_pos=None,
                key_padding_mask=None,
                reference_points=None,
                spatial_shapes=None,
                level_start_index=None,
                **kwargs):
        print(self.sampling_offsets.weight.abs().sum())
        print(self.sampling_offsets.bias.abs().sum())

        if key is None:
            key = query
        if value is None:
            value = key

        if residual is None:
            inp_residual = query
        if query_pos is not None:
            query = query + query_pos

        # change to (bs, num_query ,embed_dims)
        query = query.permute(1, 0, 2)
        value = value.permute(1, 0, 2)

        bs, num_query, _ = query.shape
        bs, num_key, _ = value.shape
        assert (spatial_shapes[:, 0] * spatial_shapes[:, 1]).sum() == num_key

        value = self.value_proj(value)
        if key_padding_mask is not None:
            value = value.masked_fill(key_padding_mask[..., None], 0.0)
        value = value.view(bs, num_key, self.num_heads, -1)
        sampling_offsets = self.sampling_offsets(query).view(
            bs, num_query, self.num_heads, self.num_levels, self.num_points, 2)
        attention_weights = self.attention_weights(query).view(
            bs, num_query, self.num_heads, self.num_levels * self.num_points)
        attention_weights = attention_weights.softmax(-1)

        attention_weights = attention_weights.view(bs, num_query,
                                                   self.num_heads,
                                                   self.num_levels,
                                                   self.num_points)
        if reference_points.shape[-1] == 2:
            offset_normalizer = torch.stack(
                [spatial_shapes[..., 1], spatial_shapes[..., 0]], -1)
            sampling_locations = reference_points[:, :, None, :, None, :] \
                                 + sampling_offsets \
                                 / offset_normalizer[None, None, None, :, None, :]
        elif reference_points.shape[-1] == 4:
            sampling_locations = reference_points[:, :, None, :, None, :2] \
                                 + sampling_offsets / self.num_points \
                                 * reference_points[:, :, None, :, None, 2:] \
                                 * 0.5
        else:
            raise ValueError(
                f'Last dim of reference_points must be'
                f' 2 or 4, but get {reference_points.shape[-1]} instead.')
        if torch.cuda.is_available():
            output = MultiScaleDeformableAttnFunction.apply(
                value, spatial_shapes, level_start_index, sampling_locations,
                attention_weights, self.im2col_step)
        else:
            output = multi_scale_deformable_attn_pytorch(
                value, spatial_shapes, level_start_index, sampling_locations,
                attention_weights, self.im2col_step)
        output = self.output_proj(output).permute(1, 0, 2)
        # (num_query, bs ,embed_dims)
        return self.dropout(output) + inp_residual

opened by LuckyDC 17

add model registry

This PR refactored Registry class in following ways:

make build function an attribute of Registry class.

before:

model = build_from_cfg(cfg, MODELS)

after:

model = MODELS.build(cfg)

build function could be configured by passing argument.

Added hierarchy structure for Registry.

For example, we can define MODEL registry in both MMDetection and MMClassification as followed: In MMDetection:

from mmcv.utils import Registry
from mmcv.cnn import MODELS as MMCV_MODELS
MODELS = Registry('model', parent=MMCV_MODELS)

@MODELS.register_module()
class NetA(nn.Module):
    def forward(self, x):
        return x

In MMClassification:

from mmcv.utils import Registry
from mmcv.cnn import MODELS as MMCV_MODELS
MODELS = Registry('model', parent=MMCV_MODELS)

@MODELS.register_module()
class NetB(nn.Module):
    def forward(self, x):
        return x + 1

We could build either NetA or NetB by:

from mmcv.cnn import MODELS as MMCV_MODELS
net_a = MMCV_MODELS.build(cfg=dict(type='mmdet.NetA'))
net_b = MMCV_MODELS.build(cfg=dict(type='mmcls.NetB'))

Modification for MM repos:

before:

from mmcv.utils import Registry, build_from_cfg
BACKBONES = Registry('backbone')
def build(cfg, registry, default_args=None):
    """Build a module.

    Args:
        cfg (dict, list[dict]): The config of modules, is is either a dict
            or a list of configs.
        registry (:obj:`Registry`): A registry the module belongs to.
        default_args (dict, optional): Default arguments to build the module.
            Defaults to None.

    Returns:
        nn.Module: A built nn module.
    """
    if isinstance(cfg, list):
        modules = [
            build_from_cfg(cfg_, registry, default_args) for cfg_ in cfg
        ]
        return nn.Sequential(*modules)
    else:
        return build_from_cfg(cfg, registry, default_args)

def build_backbone(cfg):
    """Build backbone."""
    return build(cfg, BACKBONES)

after:

from mmcv.cnn import ModelRegistry
from mmcv.cnn import MODELS as MMCV_MODELS
MODELS = ModelRegistry('models', parent=MMCV_MODELS)
def build_model(cfg):
    """Build model."""
    return MODELS.build(cfg)

opened by xvjiarui 17

Releases(v1.7.1)

v1.7.1(Jan 3, 2023)
Features

Support RoiawarePool3d, MultiScaleDeformableAttn, ModulatedDeformConv2d and NMS with cambricon MLU backend (#2359, #2396, #2411, #2401)

Add Ascend npu adapter for FusedBiasLeakyRelu, DeformableRoIPool and NMS (#2474, #2481, #2387)

Support receptive field search of CNN models (https://github.com/open-mmlab/mmcv/pull/2056)

Bug fixes

Create Tensor with new_* method to support AMP (https://github.com/open-mmlab/mmcv/pull/2389)

Fix DCN TensorRT plugin (https://github.com/open-mmlab/mmcv/pull/2408)

Fix the bug when the input mask is not '0-1-tensor' in masked_conv (https://github.com/open-mmlab/mmcv/pull/2423)

Fix UT caused by the new version of packaging (https://github.com/open-mmlab/mmcv/pull/2482)

Change the block size to 1024 during compilation (https://github.com/open-mmlab/mmcv/pull/2460)

Fix skip_layer for RF-Next (https://github.com/open-mmlab/mmcv/pull/2489)

Documentations

Update supported pytorch versions (https://github.com/open-mmlab/mmcv/pull/2382)

Add code style guide (https://github.com/open-mmlab/mmcv/pull/2406)

Add currently-npu-supported-ops in ops.md (https://github.com/open-mmlab/mmcv/pull/2420)

Refine contributing.md (https://github.com/open-mmlab/mmcv/pull/1941)

Fix typos in contributing docs (#2462, #2485)

Update qq group link (https://github.com/open-mmlab/mmcv/pull/2453)

Replace NPU with Ascend in installation documentatio (https://github.com/open-mmlab/mmcv/pull/2469)

Others

Add torch1.13 checking in CI (https://github.com/open-mmlab/mmcv/pull/2405)

Update the URL of pyupgrade pre-commit hook (https://github.com/open-mmlab/mmcv/pull/2448)

Contributors

A total of 24 developers contributed to this release. @ZhangLearning @grimoire @zhouzaida @cathyzhang222 @DerryHub @ckirchhoff2021 @wangjiangben-hw @zcc-zjut @momo609 @ZShaopeng @Wickyzheng @HAOCHENYE @gasvn @lzyhha @zytx121 @BayMaxBHL @sunyc11 @mengpenghui @RangeKing @vansin @Taited @jayggh @xiabo123 @CokeDong

New Contributors

@ZhangLearning made their first contribution in https://github.com/open-mmlab/mmcv/pull/2359

@DerryHub made their first contribution in https://github.com/open-mmlab/mmcv/pull/2408

@gasvn made their first contribution in https://github.com/open-mmlab/mmcv/pull/2056

@BayMaxBHL made their first contribution in https://github.com/open-mmlab/mmcv/pull/2448

@sunyc11 made their first contribution in https://github.com/open-mmlab/mmcv/pull/2459

@mengpenghui made their first contribution in https://github.com/open-mmlab/mmcv/pull/2411

@Taited made their first contribution in https://github.com/open-mmlab/mmcv/pull/2485

@jayggh made their first contribution in https://github.com/open-mmlab/mmcv/pull/2474

@xiabo123 made their first contribution in https://github.com/open-mmlab/mmcv/pull/2460

@lzyhha made their first contribution in https://github.com/open-mmlab/mmcv/pull/2489

Full Changelog: https://github.com/open-mmlab/mmcv/compare/v1.7.0...v1.7.1
Source code(tar.gz)
Source code(zip)
v2.0.0rc3(Nov 25, 2022)
Features

Add stack ball query and stack group points ops (#2292)

Support RoIAwarePool3d, Iou3d and MultiScaleDeformableAttn with MLU backend (#2359, #2339)

Support NMS with MLU590 backend (#2401)

Improvements

Adapt mlu code to cntoolkit3.0.1

Bug fixes

Fix roi align symbolic for torch>=1.13 (#2443)

Create Tensor with new_* method to support AMP (#2389)

Fix three nn op can not accept half tensor (#2348)

Fix the potential NaN bug in calc_square_dist (#2356)

Fix Correlation op (#2274)

Fix a not proper assertion in RandomChoiceResize (#2450)

Documentations

Make installation commands compatible with windows CMD (#2353)

Add macOS building notice (#2364)

Refine contributing.md (#1941)

Add code style guide (#2406)

Others

Add torch1.13 checking in CI (#2441)

Contributors

A total of 13 developers contributed to this release. @LYMDLUT @LYM-fire @zhouzaida @VVsssssk @xin-li-67 @grimoire @ZCMax @ZhangLearning @cathyzhang222 @HAOCHENYE @defei-coder @ZShaopeng @xiexinch

New Contributors

@LYMDLUT made their first contribution in https://github.com/open-mmlab/mmcv/pull/2353

@xiexinch made their first contribution in https://github.com/open-mmlab/mmcv/pull/2450

Full Changelog: https://github.com/open-mmlab/mmcv/compare/v2.0.0rc2...v2.0.0rc3
Source code(tar.gz)
Source code(zip)
v1.7.0(Nov 3, 2022)
Highlights

Starting from v1.7.0, we are providing pre-built packages for PyTorch v1.13.0, as well as for the macOS platform. Note that, for the mac M1 chip, only the pre-built packages on PyTorch v1.13.0 are available.

pip install -U openmim mim install mmcv-full

Features

Support training on Ascend device (https://github.com/open-mmlab/mmcv/pull/2262, https://github.com/open-mmlab/mmcv/pull/2370)

Add box_iou_quadri & nms_quadri (https://github.com/open-mmlab/mmcv/pull/2277)

Add stack ball query and stack group points ops (https://github.com/open-mmlab/mmcv/pull/2292)

Support PrRoI op for Parrots (https://github.com/open-mmlab/mmcv/pull/2280)

Support Iou3d with cambricon MLU backend (https://github.com/open-mmlab/mmcv/pull/2339)

Add getJobLimitCapability interface and use it in nms (https://github.com/open-mmlab/mmcv/pull/2337)

Improvements

Support SiLU with torch < 1.7.0 (https://github.com/open-mmlab/mmcv/pull/2278)

Update onnx registry (https://github.com/open-mmlab/mmcv/pull/2294)

Adapt mlu code to cntoolkit3.0.1 (https://github.com/open-mmlab/mmcv/pull/2248)

Refactor the nms3d op to support MLU (https://github.com/open-mmlab/mmcv/pull/2296)

Add ability to pass logger instance to frameworks (https://github.com/open-mmlab/mmcv/pull/2317)

Add a new argument define_metric in wandb hook (https://github.com/open-mmlab/mmcv/pull/2237)

Bug fixes

Fix warning of CUDA ops (https://github.com/open-mmlab/mmcv/pull/2324)

Fix the bug of uninitialized prarmeter (https://github.com/open-mmlab/mmcv/pull/2345)

Fix three nn op can not accept half tensor (https://github.com/open-mmlab/mmcv/pull/2348)

Fix Correlation op (https://github.com/open-mmlab/mmcv/pull/2274)

Fix the potential NaN bug in calc_square_dist() (https://github.com/open-mmlab/mmcv/pull/2356)

Fix MPS ops for torch1.12+ (https://github.com/open-mmlab/mmcv/pull/2302)

Documentations

Refine README and installation documentations (https://github.com/open-mmlab/mmcv/pull/1977)

Update README and fix typo in docs (https://github.com/open-mmlab/mmcv/pull/2358)

Add steps to install mmcv-full on Ascend machine (https://github.com/open-mmlab/mmcv/pull/2363)

Add maintenance plan and print deprecated message (https://github.com/open-mmlab/mmcv/pull/2357)

Others

Upgrade pre commit hooks (https://github.com/open-mmlab/mmcv/pull/2321)

Build macOS wheel in Github Action (https://github.com/open-mmlab/mmcv/pull/2311, https://github.com/open-mmlab/mmcv/pull/2366)

Contributors

A total of 14 developers contributed to this release. @okotaku @zhouzaida @grimoire @wangjiangben-hw @ckirchhoff2021 @tpoisonooo @defei-coder @budefei @CokeDong @shcheklein @cathyzhang222 @ZCMax @zytx121 @momo609

New Contributors

@tpoisonooo made their first contribution in https://github.com/open-mmlab/mmcv/pull/2324

@shcheklein made their first contribution in https://github.com/open-mmlab/mmcv/pull/2317

@cathyzhang222 made their first contribution in https://github.com/open-mmlab/mmcv/pull/2339

@momo609 made their first contribution in https://github.com/open-mmlab/mmcv/pull/2363

Full Changelog: https://github.com/open-mmlab/mmcv/compare/v1.6.2...v1.7.0
Source code(tar.gz)
Source code(zip)
v2.0.0rc2(Oct 26, 2022)
😀 We are glad to announce that we are providing the mmcv pre-built packages for macOS (x86_64 architecture).

pip install -U openmim mim install 'mmcv>=2.0.0rc2'

Features

Add TTA transform (#2146)

Add two CUDA ops, box_iou_quadri and nms_quadri (#2277)

Add two MLU ops, RoIPointPool3d and Deformable RoIPool (#2247, #2137)

Register a new activation layer SiLU to ACTIVATION_LAYERS (#2256, #2278)

Bug fixes

Fix the calculation error of out_w in MaskedConv2dFunction (#2264)

Fix MPS ops for torch1.12+ (#2302)

Add swap_labe_pairs in RandomFlip (#2332)

Documentations

Fix some mistakes in data_process.md (#2283)

Refactor the API structures of documentation (#2254)

Add a reminder in docs for macOS M1 users (#2268)

Translate data_transform.md to English (#2325)

Refine README and installation documentation (#2340)

Make installation commands compatible with Windows cmd (#2353)

Others

Update configs of CircleCI and GitHub Action (#2255)

Build macOS wheels in Github Action (#2350)

Upgrade the versions of pre-commit hooks (#2328)

Remove onnx-related files (#2295)

New Contributors

@yxzhao2022 made their first contribution in https://github.com/open-mmlab/mmcv/pull/2283

@xin-li-67 made their first contribution in https://github.com/open-mmlab/mmcv/pull/2268

@LYMDLUT made their first contribution in https://github.com/open-mmlab/mmcv/pull/2353

Contributors

@yxzhao2022 @LYMDLUT @grimoire @xin-li-67 @zytx121 @MeowZheng @HAOCHENYE @zhouzaida @ZShaopeng @defei-coder @okotaku

Full Changelog: https://github.com/open-mmlab/mmcv/compare/v2.0.0rc1...v2.0.0rc2
Source code(tar.gz)
Source code(zip)
v1.6.2(Sep 29, 2022)
Features

Register a new activation layer SiLU to ACTIVATION_LAYERS (#2256)

Add support of TorchVision's Model Registration API (#2246)

Add five MLU ops, which are CARAFEcarafe, MaskedConv2d, ThreeNN, DeformRoiPool, and RoipointPool3d (#2212, #2202, #2215, #2137,#2247)

Improvements

Add support for logging parameters to MlflowLoggerHook (#2188)

Add a new argument params to MlflowLoggerHook (#2193)

Support different border modes in imrotate (#2141)

Add pillow backend for ColorJitter related functions (#2127)

Bug fixes

Fix unit tests of onnx (#2155)

Fix cumulative gradient optimizer when resuming (#2093)

Path mapping only once in storage backends (#2204)

Fix RoiAlignRotated MLU implementation (#2210)

Fix a potential bug in PrRoIPool op (#2200)

Fix spconv clang-format (#2260)

Remove the restrictions of onnx versions (#2251)

Fix init_weights of MultiScaleDeformableAttention (#2158)

Cast the type of mask in ModulatedDeformConv2dFunction to enable training with AMP (#2220)

Fix the calculation error of out_w in MaskedConv2dFunction (#2264)

Fix ROCm backend support (#1918)

Documentations

Limit sphinx_markdown_tables and markdown versions (#2144)

Add swith_language.md in docs (#2160)

Fix the example error in registry.md (#2243)

Add zhihu articles (#2157)

Fix some mistakes in data_process.md (#2283)

Others

Add type hints for mmcv/image (#2089)

New Contributors

@gfyts0123 made their first contribution in https://github.com/open-mmlab/mmcv/pull/2103

@xcnick made their first contribution in https://github.com/open-mmlab/mmcv/pull/2155

@eivtho made their first contribution in https://github.com/open-mmlab/mmcv/pull/2188

@CokeDong made their first contribution in https://github.com/open-mmlab/mmcv/pull/2189

@tudejiang79 made their first contribution in https://github.com/open-mmlab/mmcv/pull/2210

@DanieeelLiu made their first contribution in https://github.com/open-mmlab/mmcv/pull/2215

@PungTeng made their first contribution in https://github.com/open-mmlab/mmcv/pull/2243

@ZShaopeng made their first contribution in https://github.com/open-mmlab/mmcv/pull/2247

@zstreet87 made their first contribution in https://github.com/open-mmlab/mmcv/pull/1918

@okotaku made their first contribution in https://github.com/open-mmlab/mmcv/pull/2256

@datumbox made their first contribution in https://github.com/open-mmlab/mmcv/pull/2246

Full Changelog: https://github.com/open-mmlab/mmcv/compare/v1.6.1...v1.6.2
Source code(tar.gz)
Source code(zip)
v2.0.0rc1(Aug 31, 2022)

Highlights

Starting from mmcv v2.0.0rc1, mmcv removes training process related modules such as Runner, Hook, Parallel, which will be provided by MMEngine. In addition, in v2.0.0rc1, mmcv added many data augmentation methods, such as CenterCrop.
Source code(tar.gz)
Source code(zip)
v1.6.1(Jul 26, 2022)
😀 We are glad to announce that we are providing the mmcv-full pre-built packages for PyTorch 1.12.0 starting from v1.6.0.

pip install -U openmim mim install mmcv-full

Features

Add bbox overlap MPS op (#2123)

Add PrRoIPool CUDA op (#2113)

Improvements

Improve ProfilerHook behavior when using a relative (or empty) dir_name (#2119)

Documentations

Fix some mistakes of examples (#2120)

Fix typos (#2124)

Fix markdown version (#2121)

Others

Add .pre-commit-config-zh-cn.yaml to speed up pre-commit in some cases

Contributors

A total of 6 developers contributed to this release.

@JingweiZhang12 @grimoire @zhouzaida @GPhilo @0reza @Happylkx

New Contributors

@Happylkx made their first contribution in https://github.com/open-mmlab/mmcv/pull/2120

@0reza made their first contribution in https://github.com/open-mmlab/mmcv/pull/2124

@GPhilo made their first contribution in https://github.com/open-mmlab/mmcv/pull/2119

@JingweiZhang12 made their first contribution in https://github.com/open-mmlab/mmcv/pull/2113

Full Changelog: https://github.com/open-mmlab/mmcv/compare/v1.6.0...v1.6.1
Source code(tar.gz)
Source code(zip)
v1.6.0(Jul 12, 2022)
😀 We are glad to announce that we are providing the mmcv-full pre-built packages for PyTorch 1.12.0 starting from v1.6.0.

pip install -U openmim mim install mmcv-full

Features

Add support for MPS (#2092)

Support RoiPool with Cambricon MLU backend (#2073)

Improvements

Add torch_meshgrid wrapper due to PyTorch change (#2044)

Add _run_ddp_forward method for MMDistributedDataParallel to be compatible with PyTorch1.12 (#2107)

Bug fixes

Fix name error in trt_deform_conv.cpp (#2075)

Add torchvision_0.12.json for testing torchvision0.13 (#2084)

Fix num_channels in test_build_norm_layer which can not be divisible by num_groups (#2085)

Fix index error when using multi-samplers strategy (#2094)

Fix undefined variable error in pixel_group.cpp (#2104)

Type Hints

#2031 #1991 #2003 #1995 #1993 #2030 #2002

Documentations

Reformat the op docs (#2052)

Add Dockerfile (#1974)

Contributors

A total of 14 developers contributed to this release.

@imabackstabber @triple-Mu @HAOCHENYE @zhouzaida @ZwwWayne @pallgeuer @triwahyuu @ytzhao @WINDSKY45 @EvelynWang-0423 @grimoire @zcyKTH @zhangtingyu11 @ychfan
Source code(tar.gz)
Source code(zip)
v1.5.3(Jun 15, 2022)
Features

Add ChamferDistance CUDA op (#1933)

Add two MLU ops, RoiAlignRotated and PSAMask (#2033 #2024)

Bug fixes

Fix protobuf compatible error (#1992)

Include torchvision_12.json in the sdist package (#2039)

make iou3d.boxes_iou3d actually calculate 3D IoU (#2018)

Fix batched_nms for rotated box and add type hints for nms.py (#2006)

Type Hints

Thanks to contributors for their efforts, mmcv has added type hints, which makes the code more readable. Here are all the pull requests of adding type hints.

#1981 #1989 #1998 #1985 #1987 #1983 #2005 #1988 #2013 #2007 #2015 #2017 #2001 #2019 #2021 #2023 #2000 #2026 #1997 #2029 #2016 #2008 #2031 #1982 #1994 #1999 #2020 #2037 #2034 #1996 #2050

Documentations

Update copyright years in docs (#2042)

Fix mdformat version to support python3.6 (#2014)

Contributors

A total of 22 developers contributed to this release.

@RangeKing @gy-7 @triple-Mu @ytzhao @zengxiang68 @yangrisheng @WINDSKY45 @ShawnHXH @vansin @HarborYuan @tcexeexe @luopeichao @zcyKTH @filaPro @nxznm @liuyanyi @grimoire @zhouzaida @HAOCHENYE @ZwwWayne @teamwong111 @imabackstabber
Source code(tar.gz)
Source code(zip)
v1.5.2(May 30, 2022)
Improvements

Lower the restrictions of _resize method in BaseMergeCell (#1959)

Replace BEV IoU with 3D IoU (#1902)

Hook can get data_batch from Runner (#1940)

Add pyupgrade pre-commit hook (#1937)

Bug fixes

Fix bbox overlap fp16 (#1958)

Skip CPU test in test_rotated_feature_align.py for parrots (#1968)

Fix is_module_wrapper (#1900)

Fix the problem that the instance cannot be reused when overwriting the file client (#1747)

Documentations

Replace markdownlint with mdformat for avoiding installing ruby (#1936)

Fix failed links in docs/zh_cn/faq.md (#1964)

Fix docstrings in EvalHook (#1978)

Notes

Since PyTorch 1.11.0 has been released and we have provided the mmcv-full pre-compiled packages for it, we no longer provide them compiled with PyTorch 1.3 & 1.4 since v1.3.17. You can find previous versions compiled with PyTorch 1.3 & 1.4 here. The compatibility is still ensured in our CI, but we will discard the support of PyTorch 1.3 & 1.4 this year.

Contributors

A total of 14 developers contributed to this release.

@grimoire @teamwong111 @zhouzaida @JarvisKevin @HAOCHENYE @jshilong @filaPro @HIT-cwh @hhaAndroid @ZwwWayne @Tai-Wang @274869388 @wlf-darkmatter @imabackstabber
Source code(tar.gz)
Source code(zip)
v1.5.1(May 17, 2022)
Features

Added ClearMLLoggerHook (#1906)

Added rotated_feature_align cpu & onnxruntime implementations (#1878)

Supports registering function (#1858)

Added adjust_hue (#1628)

Added a script to visualize lr (#1667)

Improvements

Speeds up bbox_overlaps op (#1718)

Added support for latest Pillow resampling filter naming scheme (#1931)

Bug fixes

Fixed collec_env() encoding error on Windows (#1905)

Fixed dump method of Config (#1837)

Fixed bbox_overlaps unit tests (#1914)

Ensures the type of filename parameter in Config is str (#1725)

Fixed nms rotate illegal memory access (#1891)

Fixed LmdbBackend which could not pickle env obj (#1764)

Fixed type annotations in file_client (#1942)

Fixed bbox overlap fp16 (#1958)

Documentations

Added IPU installation documentation (#1904)

Added type annotations (#1947, #1950, #1946)

Contributors

A total of 18 developers contributed to this release.

@imabackstabber @akravchuk97 @SuTanTank @grimoire @teamwong111 @HuDi2018 @WilliamKyle @luopeichao @274869388 @bryandeng @tehkillerbee @cuhk-hbsun @jmercat @triple-Mu @sundongwei @teamwong111 @ZwwWayne @zhouzaida
Source code(tar.gz)
Source code(zip)
v1.5.0(Apr 21, 2022)
Features

Support training on IPU chip (#1882)

Support training on MLU chip (#1770, #1563,#1884,#1885,#1881)

Support for linearly learning rate decay (#1627)

Improvements

Optimize correlation (#1814)

Speed up Registry initialization (#1844)

NMS with CUDA only (#1824)

Add a differentiable rotated IoU op (#1854)

Support passing None to DictAction (#1834)

Enhance get_torchvision_models (#1867)

Bug fixes

Fix Python3.10 CI (#1828)

Fix collect_env() in Windows (#1789, #1887)

Fix sparse_conv unit test by releasing cache on GPU (#1872)

Fix unexpected keyword argument for GELU (#1847)

Fix potential negative padding size (#1792)

Fix version comparison to include prerelease versions (#1877)

Documentations

Fix docstring (#1876, #1880)

Notes

Since PyTorch 1.11.0 has been released and we have provided the mmcv-full pre-compiled packages for it, we no longer provide them compiled with PyTorch 1.3 & 1.4 since v1.3.17. You can find previous versions compiled with PyTorch 1.3 & 1.4 here. The compatibility is still ensured in our CI, but we will discard the support of PyTorch 1.3 & 1.4 this year.

Contributors

A total of 29 developers contributed to this release.

@wulele2 @Sharpiless @RockeyCoss @teamwong111 @SuTanTank @grimoire @MeowZheng @triple-Mu @astahlman @Qiza-lyhm @daydayfun @VVsssssk @Charlyo @filaPro @zytx121 @DCNSW @luopeichao @gaotongxiao @yushinliu @zihanchang11 @raoshenglong @zcyKTH @Mrxiaofei @HuDi2018 @vakker @zhouzaida @ZwwWayne @HAOCHENYE @imabackstabber
Source code(tar.gz)
Source code(zip)
v1.4.8(Mar 29, 2022)
😀 We are glad to announce that we are providing the mmcv-full pre-built packages for torch1.11.0 starting from v1.4.7.

# pip install mmcv-full -f https://download.openmmlab.com/mmcv/dist/{cu_version}/{torch_version}/index.html pip install mmcv-full -f https://download.openmmlab.com/mmcv/dist/cu115/torch1.11/index.html pip install mmcv-full -f https://download.openmmlab.com/mmcv/dist/cu113/torch1.11/index.html # PyTorch does not provide pre-built packages for cu102 on Windows, so does mmcv-full pip install mmcv-full -f https://download.openmmlab.com/mmcv/dist/cu102/torch1.11/index.html

Features

Config ignores imported modules and functions (#1802)

Implemented corner pool with python for torch<1.5 (#1772)

Improvements

Removed warnings in compilation (#1675)

Updated unit tests of mmdet3d ops (#1811)

Bug fixes

Fix ROCm support (#1704)

Fix MultiScaleDeformableAttention (#1823)

Fix voxelization bug (#1822)

Fix error cased by _sync_params which was renamed in torch1.11.0 (#1816)

Documentations

Add more Windows information in README (#1829)

Notes

Since PyTorch 1.10.0 has been released and we have provided the mmcv-full pre-compiled packages for it, we no longer provide them compiled with PyTorch 1.3 & 1.4 since v1.3.17. You can find previous versions compiled with PyTorch 1.3 & 1.4 here. The compatibility is still ensured in our CI, but we will discard the support of PyTorch 1.3 & 1.4 this year.

Contributors

A total of 12 developers contributed to this release.

@SuTanTank @VVsssssk @wHao-Wu @grimoire @jshilong @zhanggefan @luopeichao @imabackstabber @HAOCHENYE @zhouzaida @ZwwWayne @teamwong111
Source code(tar.gz)
Source code(zip)
v1.4.7(Mar 21, 2022)
😀 We are glad to announce that we are providing the mmcv-full pre-built packages for torch1.11.0 starting from v1.4.7.

# pip install mmcv-full -f https://download.openmmlab.com/mmcv/dist/{cu_version}/{torch_version}/index.html pip install mmcv-full -f https://download.openmmlab.com/mmcv/dist/cu115/torch1.11/index.html pip install mmcv-full -f https://download.openmmlab.com/mmcv/dist/cu113/torch1.11/index.html # PyTorch does not provide pre-built packages for cu102 on Windows, so does mmcv-full pip install mmcv-full -f https://download.openmmlab.com/mmcv/dist/cu102/torch1.11/index.html

Features

Add SegmindLoggerHook (#1650)

Config supports shallow copy (#1796)

Support fp16 training for gather_points op (#1758)

Add non-deterministic voxelization op from mmdet3d (#1783)

Add worker_init_fn helper function to initialize dataloader workers (#1788)

Bug fixes

Fix deconv_flops_counter_hook (#1760)

Fix deform_conv inconsistent pybind (#1781)

Fix ProfilerHook when using IterBasedRunner (#1755)

Include spconv header files in the sdist package (#1766)

Fix bugs in DynamicScatter and Voxelization ops (#1748, #1746)

Documentations

Fix typos in config.md, cnn.md and data_process.md (#1797, #1798, #1799)

Fix a wrong example in DefaultOptimizerConstructor (#1795)

Deprecations

Add deprecated warning of deployment (#1750)

Others

Add copyright pre-commit-hook (#1742)

Reorganize OpenMMLab projects in README.md (#1744)

Fix typo in conv.py (#1759)

Add .owners.yml to dispatch daily issue (#1775)

Add PyTorch1.11.0 CI (#1800)

Notes

Since PyTorch 1.10.0 has been released and we have provided the mmcv-full pre-compiled packages for it, we no longer provide them compiled with PyTorch 1.3 & 1.4 since v1.3.17. You can find previous versions compiled with PyTorch 1.3 & 1.4 here. The compatibility is still ensured in our CI, but we will discard the support of PyTorch 1.3 & 1.4 this year.

Contributors

A total of 15 developers contributed to this release.

@saurbhc @XiwuChen @Xiangxu-0103 @SuTanTank @imflash217 @wangruohui @imabackstabber @del-zhenwu @gaotongxiao @HAOCHENYE @zhouzaida @RunningLeon @ZwwWayne @wHao-Wu @teamwong111
Source code(tar.gz)
Source code(zip)
v1.4.6(Feb 26, 2022)
😀 We are glad to announce that we are providing the mmcv pre-built packages on Windows starting from v1.4.0.

# pip install mmcv-full -f https://download.openmmlab.com/mmcv/dist/{cu_version}/{torch_version}/index.html pip install mmcv-full -f https://download.openmmlab.com/mmcv/dist/cu102/torch1.10.0/index.html # We can ignore the micro version of PyTorch pip install mmcv-full -f https://download.openmmlab.com/mmcv/dist/cu102/torch1.10/index.html

Features

Add detection3d op spconv (#1581)

Support deepcopy for Config (#1658)

Improvements

Refactor the interface of RoIAlignRotated (#1662)

Replace positional argument with keyword argument in EvalHook when calling save_checkpoint (#1694)

Bug fixes

Remove unused header files (#1653)

Documentations

Fix spelling error in registry.md (#1717)

Notes

Since PyTorch 1.10.0 has been released and we have provided the mmcv-full pre-compiled packages for it, we no longer provide them compiled with PyTorch 1.3 & 1.4 since v1.3.17. You can find previous versions compiled with PyTorch 1.3 & 1.4 here. The compatibility is still ensured in our CI, but we will discard the support of PyTorch 1.3 & 1.4 this year.

Contributors

A total of 12 developers contributed to this release.

@rentainhe @mzr1996 @MeowZheng @wHao-Wu @grimoire @ZwwWayne @AllentDan @teamwong111 @zhouzaida @nijkah @luopeichao @makifozkanoglu
Source code(tar.gz)
Source code(zip)
v1.4.5(Feb 15, 2022)
😀 We are glad to announce that we are providing the mmcv pre-built packages on Windows starting from v1.4.0.

# pip install mmcv-full -f https://download.openmmlab.com/mmcv/dist/{cu_version}/{torch_version}/index.html pip install mmcv-full -f https://download.openmmlab.com/mmcv/dist/cu102/torch1.10.0/index.html # We can ignore the micro version of PyTorch pip install mmcv-full -f https://download.openmmlab.com/mmcv/dist/cu102/torch1.10/index.html

Improvements

Replace HSwish with PyTorch implementation (#1709)

Bug fixes

Fix the undefined name error (#1695)

Fix an incorrect assignment in the BaseTransformerLayer (#1670)

Fix the mismatch torch version of MlflowLoggerHook (#1680)

Fix a bug in min_area_polygons (#1702)

Fix batched nms for nms_rotated (#1688)

Documentations

Add cuDNN installation steps in tensorrt_plugin.md (#1711)

Notes

Since PyTorch 1.10.0 has been released and we have provided the mmcv-full pre-compiled packages for it, we no longer provide them compiled with PyTorch 1.3 & 1.4 since v1.3.17. You can find previous versions compiled with PyTorch 1.3 & 1.4 here. The compatibility is still ensured in our CI, but we will discard the support of PyTorch 1.3 & 1.4 this year.

Contributors

A total of 11 developers contributed to this release.

@nijkah @zytx121 @jamiechoi1995 @jshilong @MeowZheng @RangiLyu @RunningLeon @HAOCHENYE @ZwwWayne @zhouzaida @grimoire
Source code(tar.gz)
Source code(zip)
v1.4.4(Jan 26, 2022)
😀 We are glad to announce that we are providing the mmcv pre-built packages on Windows starting from v1.4.0.

# pip install mmcv-full -f https://download.openmmlab.com/mmcv/dist/{cu_version}/{torch_version}/index.html pip install mmcv-full -f https://download.openmmlab.com/mmcv/dist/cu102/torch1.10.0/index.html # We can ignore the micro version of PyTorch pip install mmcv-full -f https://download.openmmlab.com/mmcv/dist/cu102/torch1.10/index.html

Features

Add 6 rotated detection ops for parrots (#1665)

Gradually reduce the difference between max_lr and base_lr of CyclicLrUpdaterHook (#1655)

Improvements

Improve the docstring of LoggerHooks (#1664)

Bug fixes

Fix the unit test of correlation op (#1659)

Fix the invalid CUDA memory access when training s2anet (#1660)

Fix the bug of momentum updater with warmup policy (#1626)

Align HSigmoid with PyTorch official implementation(#1622)

Fix the scatter function when the device type of input is CPU (#1621)

Fix the cast_tensor_type (#1639)

Notes

Since PyTorch 1.10.0 has been released and we have provided the mmcv-full pre-compiled packages for it, we no longer provide them compiled with PyTorch 1.3 & 1.4 since v1.3.17. You can find previous versions compiled with PyTorch 1.3 & 1.4 here. The compatibility is still ensured in our CI, but we will discard the support of PyTorch 1.3 & 1.4 this year.

Contributors

A total of 11 developers contributed to this release.

@hokmund @HAOCHENYE @ZwwWayne @teamwong111 @zhouzaida @luopeichao @Bilibilee @gaotongxiao @imyhxy @zhanggefan @grimoire
Source code(tar.gz)
Source code(zip)
v1.4.3(Jan 13, 2022)
😀 We are glad to announce that we are providing the mmcv pre-built packages on Windows starting from v1.4.0.

# pip install mmcv-full -f https://download.openmmlab.com/mmcv/dist/{cu_version}/{torch_version}/index.html pip install mmcv-full -f https://download.openmmlab.com/mmcv/dist/cu102/torch1.10.0/index.html

Features

Add three rotated detection ops, which are ConvexIoU, ActiveRotatedFilter and MinAreaPolygon (#1610, #1598, #1611)

Add artifact logging to wandb hook (#1616)

Improvements

Deprecate the support for "python setup.py test" (#1637)

Add clockwise argument to ops box_iou_rotated and nms_rotated (#1592)

Replace DIVUP with GET_BLOCKS (#1586)

Refactor requirements (#1536)

Bug fixes

Fix the compilation error on windows (#1657, #1643)

Fix iou3d bug in parrots (#1656)

Documentations

Add the link of mmdeploy (#1638)

Notes

Since PyTorch 1.10.0 has been released and we have provided the mmcv-full pre-compiled packages for it, we no longer provide them compiled with PyTorch 1.3 & 1.4 since v1.3.17. You can find previous versions compiled with PyTorch 1.3 & 1.4 here. The compatibility is still ensured in our CI, but we will discard the support of PyTorch 1.3 & 1.4 next year.

Contributors

A total of 12 developers contributed to this release.

@zytx121 @fcakyon @zhanggefan @LudoBar @zhouzaida @ZwwWayne @grimoire @luopeichao @MeowZheng @AllentDan @Johnson-Wang @teamwong111
Source code(tar.gz)
Source code(zip)
v1.4.2(Dec 25, 2021)
😀 We are glad to announce that we are providing the mmcv pre-built packages on Windows starting from v1.4.0.

# pip install mmcv-full -f https://download.openmmlab.com/mmcv/dist/{cu_version}/{torch_version}/index.html pip install mmcv-full -f https://download.openmmlab.com/mmcv/dist/cu102/torch1.10.0/index.html

Features

Add three rotated detection ops, which are RotatedFeatureAlign, RiRoIAlignRotated and PointsInPolygons (#1603, #1599, #1600)

Support device dispatch in parrots (#1588)

Improvements

Fix some warnings in the unit tests (#1522)

Improve the DvcliveLoggerHook (#1208)

tensor2imgs supports converting a tensor to gray image (#1595)

Bug fixes

Compute distance to origin after sort (#1500)

Convert the type of bias to the same as the input in ModulatedDeformConv2dFunction (#1605)

Documentations

Refactor the structure of documentation (#1580)

Fix the typo in runner.md (#1591)

Notes

Since PyTorch 1.10.0 has been released and we have provided the mmcv-full pre-compiled packages for it, we no longer provide them compiled with PyTorch 1.3 & 1.4 since v1.3.17. You can find previous versions compiled with PyTorch 1.3 & 1.4 here. The compatibility is still ensured in our CI, but we will discard the support of PyTorch 1.3 & 1.4 next year.

Contributors

A total of 11 developers contributed to this release.

@teamwong111 @grimoire @uniyushu @daavoo @cuhk-hbsun @zhouzaida @ZwwWayne @gaotongxiao @AceCoooool @zytx121 @luopeichao
Source code(tar.gz)
Source code(zip)
v1.4.1(Dec 16, 2021)
😀 We are glad to announce that we are providing the mmcv pre-built packages on Windows starting from v1.4.0.

# pip install mmcv-full -f https://download.openmmlab.com/mmcv/dist/{cu_version}/{torch_version}/index.html pip install mmcv-full -f https://download.openmmlab.com/mmcv/dist/cu102/torch1.10.0/index.html

Features

Add ModuleDict inherited from BaseModule and ModuleDict (#1542)

Add PatchEmbed and PatchMerging modules used by transformer methods (#1305)

Add multiple backends support for imread/imwrite (#1527)

Improvements

Cancel previous CI jobs that are not completed or in the queue (#1537)

Detect anomalous parameters for debugging (#1547)

batched_nms supportd skipping the NMS operation (#1552)

Bug fixes

Fix the TensorRT error with PyTorch 1.8+ (#1464)

Fix _get_checkpoint_loader that can not correctly obtain a loader when the path of ceph contains cluster name (#1577)

Fix load_from_local that can not handle the path starting with ~ (#1549)

Fix the type inconsistency when a config contains the _delete_ key (#1575)

Documentations

Fix the documentation of deployment (#1512)

Fix the format of docstring (#1462, #1573)

Update FAQ (#1481)

Notes

Since PyTorch 1.10.0 has been released and we have provided the mmcv-full pre-compiled packages for it, we no longer provide them compiled with PyTorch 1.3 & 1.4 since v1.3.17. You can find previous versions compiled with PyTorch 1.3 & 1.4 here. The compatibility is still ensured in our CI, but we will discard the support of PyTorch 1.3 & 1.4 next year.

Contributors

A total of 10 developers contributed to this release.

@lvhan027 @RunningLeon @grimoire @zhouzaida @ZwwWayne @LutingWang @MeowZheng @teamwong111 @twmht @274869388
Source code(tar.gz)
Source code(zip)
v1.4.0(Dec 3, 2021)
😀 We are glad to announce that we are providing the mmcv pre-built packages on Windows starting from v1.4.0.

# pip install mmcv-full -f https://download.openmmlab.com/mmcv/dist/{cu_version}/{torch_version}/index.html pip install mmcv-full -f https://download.openmmlab.com/mmcv/dist/cu102/torch1.10.0/index.html

Improvements

Replace the third-party markdownlint hook with the official (#1535)

Add the CircleCI to test the unit tests of CUDA (#1540)

Skip CI when some specific files were changed (#1533)

Bug fixes

Fix the python bind of iou3d (#1524)

Fix a compiled error on Windows (#1526, #1543)

Fix a compiled error on Linux with cuda9.0 (#1525)

Fix an undefined symbol error when installing from pip (#1544)

Notes

Since PyTorch 1.10.0 has been released and we have provided the mmcv-full pre-compiled packages for it, we no longer provide them compiled with PyTorch 1.3 & 1.4 since v1.3.17. You can find previous versions compiled with PyTorch 1.3 & 1.4 here. The compatibility is still ensured in our CI, but we will discard the support of PyTorch 1.3 & 1.4 next year.

Contributors

A total of 5 developers contributed to this release.

@grimoire @teamwong111 @zhouzaida @ZwwWayne @wangruohui @del-zhenwu
Source code(tar.gz)
Source code(zip)
v1.3.18(Nov 25, 2021)
New features

Add group_points, iou3d, roiaware_pool3d and voxelize in parrots (#1504)

Improvements

Refactor csrc with device dispatcher (#1463)

Update MMCls models in model_zoo/mmcls.json (#1513)

Improve .gitignore about datasets and logs and checkpoints (#1477)

Bug fixes

Fix compiled error on Windows (#1510)

Fix the unit test of Deform Conv (#1490)

Fix train example (#1502)

Fix ops links in LICENSES (#1517)

Skip the checking of LOCAL_RANK when loading checkpoint from HTTP or HTTPS (#1396)

Add load_url to handle incompatibility of PyTorch versions (#1377)

Docs

Add a script to check the installation of mmcv-full (#1474)

Add the URL of pre-built packages built with CUDA 11.3 (#1489)

Fix grammatical error in Registry docs (#1516)

Add the link and introduction of MMFlow in README (#1506)

Notes

Since PyTorch 1.10.0 has been released and we have provided the mmcv-full pre-compiled packages for it, we no longer provide them compiled with PyTorch 1.3 & 1.4 since v1.3.17. You can find previous versions compiled with PyTorch 1.3 & 1.4 here. The compatibility is still ensured in our CI, but we will discard the support of PyTorch 1.3 & 1.4 next year.

Contributors

A total of 13 developers contributed to this release.

@grimoire @lzhangzz @RunningLeon @zhouzaida @ZwwWayne @teamwong111 @MeowZheng @luopeichao @mzr1996 @vansin @tobiasfshr @innerlee @ly015
Source code(tar.gz)
Source code(zip)
v1.3.17(Nov 11, 2021)
Improvements

Improve the log when loading checkpoints (#1446)

Bug fixes

Fix the wrong weight reference in BaseTransformerLayer (#1418)

Sleep 2 seconds to prevent deadlock in IterBasedRunner (#1442)

Fix deform conv by adding an extra argument im2col_step (#1459)

Print a warning log if eval_res is an empty dict (#1398)

Rename the init parameter task to name in PaviLoggerHook (#1438)

Update the test data of test_tin_shift (#1426)

Docs

Add ops definition (#1434)

Fix invalid symbols in docstring (#1466)

Update the table of pre-built packages in README (#1433)

Notes

Since PyTorch 1.10.0 has been released and we have provided the mmcv-full pre-compiled packages for it, we no longer provide them compiled with PyTorch 1.3 & 1.4 since v1.3.17. You can find previous versions compiled with PyTorch 1.3 & 1.4 here. The compatibility is still ensured in our CI, but we will discard the support of PyTorch 1.3 & 1.4 next year.

Contributors

A total of 12 developers contributed to this release.

@grimoire @gaotongxiao @zhouzaida @ZwwWayne @teamwong111 @MeowZheng @ly015 @nbei @RangiLyu @Junjun2016 @hhaAndroid @RunningLeon
Source code(tar.gz)
Source code(zip)
v1.3.16(Oct 25, 2021)
New features

Add 4 detection3d related CUDA operators, which are voxel op, roiaware pool3d op, iou3d op and group points op (#1381, #1382, #1356, #1415)

Power FileClient and make it support more file interfaces (#1330)

Support uploading (loading) checkpoints to (from) different backends (#1375)

Improvements

Add CI for PyTorch 1.10 (#1431)

Add case_sensitive argument in scandir (#1389)

Bug fixes

Add DeviceGuard for ms_deform_attn_forward (#1402)

Fix config parsing error caused by non-ascii characters (#1410)

Fix error when building pdf documentaiton (#1414)

Update test data for test_iou3d (#1427)

Docs

Polish the documentation (#1411)

Notes

Since PyTorch 1.10.0 has been released and we have provided the pre-compiled packages for it, we plan to end the support for PyTorch 1.3 and 1.4 in future releases. We will still provide pre-compiled packages of these two versions for three months, but the reliability and technical support are not guaranteed.

Contributors

A total of 11 developers contributed to this release.

@jshilong , @bobo0810 , @ly015 , @zhouzaida , @Ezra-Yu , @DCNSW, @grimoire , @luopeichao , @ZwwWayne , @nbei , @mzr1996
Source code(tar.gz)
Source code(zip)
v1.3.15(Oct 16, 2021)
New features

Add seven detection3d related CUDA operators, which are FurthestPointSample, KNN, GatherPoints, ThreeInterpolate, ThreeNN, AssignScoreWithK and RoIPointPool3d (#1337, #1354, #1338, #1355, #1357, #1358, )

Add Deformable Conv CustomOp for onnxruntime (#1343)

Improvements

Switch the CUDA CI to the docker image provided by PyTorch (#1376)

Add code spell hook to pre-commit and fix typos (#1384)

Remove some obsolete code (#1370)

Bug fixes

Fix the weight initialization of deform conv (#1262)

Revise the unit test of correlation (#1368)

Fix the unrecognized symbol in correlation.cpp (#1372)

Fix a potential bug in deprecated_api_warning (#1395)

Add error handler when getting host failed (#1374)

Fix the infernence issue of MultiScaleDeformableAttention on CPU device (#1223)

Add a condition to disable invalid memory write by furthest_sample_point (#1405)

Reduce the test memory usage of ms_deformable_attn (#1407)

Docs

Add the Chinese documentation for CNN, ONNX, TensorRT, Build mmcv-full from source and contributing (#1246, #1331, #1336, #1353, #1351)

Add a note for installing mmcv-full (#1320)

Notes

Since PyTorch 1.9 has been released, we plan to end the support for PyTorch 1.3 and 1.4 in future releases. We will still provide pre-compiled packages of these two versions for three months, but the reliability and technical support are not guaranteed. In addition, starting from this version, we provide pre-compiled packages for Python 3.9.

Contributors

A total of 12 developers contributed to this release.

@bofen97, @grimoire, @zhouzaida, @ZwwWayne, @MeowZheng, @Fu0511, @hhaAndroid, @eugene123tw, @DCNSW, @del-zhenwu, @ZhichengHuang, @jshilong
Source code(tar.gz)
Source code(zip)
v1.3.14(Sep 24, 2021)
New features

Add two CUDA OPs BallQuery and Correlation (#1332, #1361 )

Support loading optical flow data from bytes (#1362)

Support empty tensor in MMSyncBN (#1205)

Docs

Replace the default theme of documentation with PyTorch Sphinx Theme (#1321)

Add the Chinese documentation for registry.md, config.md, ops.md and visualization.md (#1204, #1349, #1350)

Notes

Since PyTorch 1.9 has been released, we plan to end the support for PyTorch 1.3 and 1.4 in future releases. We will still provide pre-compiled packages of these two versions for three months, but the reliability and technical support are not guaranteed.

Contributors

A total of 8 developers contributed to this release.

@AluminiumOxide, @MeowZheng, @zhouzaida, @grimoire, @ZwwWayne, @nbei, @DCNSW, @luopeichao
Source code(tar.gz)
Source code(zip)
v1.3.13(Sep 10, 2021)
New features

Add the CPU implementation of DCN and Modulated DCN (#1278)

Add the revert_sync_batchnorm function to convert SyncBN in any model to BN (#1253)

Add python3.9 in CI to verify the support for python3.9 (#1291)

Improvements

Reimplement cc_attention using pure PyTorch (#1201)

Enhance the output of exception (#1309)

Restrict the warning message (#1267)

Bug fixes

Fix missing state_dict._metadata when saving and loading checkpoints (#1294)

Fix typo about installation (#1313)

Fix sphinx version (#1301)

Docs

Add the Chinese documentation for faq.md (#1252)

Notes

Since PyTorch 1.9 has been released, we plan to end the support for PyTorch 1.3 and 1.4 in future releases. We will still provide pre-compiled packages of these two versions for three months, but the reliability and technical support are not guaranteed.

Contributors

A total of 13 developers contributed to this release.

@Leojc @eugene123tw @weiji14 @mzr1996 @zhouzaida @yyz561 @gaotongxiao @EricKani @MeowZheng @ZwwWayne @xvjiarui @Junjun2016 @grimoire
Source code(tar.gz)
Source code(zip)
v1.3.12(Aug 24, 2021)
New features

Add a GradientCumulativeOptimizerHook for training model with a larger batch size (#1221)

Add a DefaultRunnerConstructor for extending existing Runner(#1296)

Add modulated_deform_conv to support onnxruntime (#1281)

Improvements

Add mmcv namespace for custom onnx ops (#1254)

Compile mmcv with all available CPUs (#1046)

Support deprecation information in Config (#1275)

Add windows CI (#1023)

Bug fixes

Fix the nms implementation of CPU (#1244)

Remove the duplicate files in mmcv/ops/csrc (#1284)

Upgrade the version of ONNX Runtime (#1290)

Docs

Use myst-parser extension for sphinx and make the style more beautiful (#1287)

Add the Chinese documentation for utils.md and runner.md (#1232, #1202)

Notes

Since PyTorch 1.9 has been released, we plan to end the support for PyTorch 1.3 and 1.4 in future releases. We will still provide pre-compiled packages of these two versions for three months, but the reliability and technical support are not guaranteed.

Contributors

A total of 11 developers contributed to this release.

@innerlee, @grimoire, @zhouzaida, @raoshenglong, @jiang1997, @ZhiyuanChen, @mzr1996, @PeiqiWang, @hhaAndroid, @ZwwWayne, @densechen
Source code(tar.gz)
Source code(zip)
v1.3.11(Aug 12, 2021)
Improvements

Refactor the directory of csrc (#1206)

Add PR document (#1209)

Speed up mmcv import (#1249)

Do not depend on setuptools>=52 (#1235)

Bug Fixes

Fix the bug that the training log and evaluating log are mixed (#1252)

Fix dcn forward and backward when batch size is larger than im2col_step (#1212)

Fix potential bugs of basemodule when record the initilization information (#1217)

Notes

Since PyTorch 1.9 has been released, we plan to end the support for PyTorch 1.3 and 1.4 in the future releases. We will still provide pre-compiled packages of these two versions for three months, but the reliability and technical support are not guaranteed.

Contributors

A total of 12 developers contributed to this release.

@hellock, @xvjiarui, @jshilong, @ZwwWayne, @MeowZheng, @innerlee, @zhouzaida, @kennymckormick, @grimoire, @nbei, @dreamerlin, @Junjun2016
Source code(tar.gz)
Source code(zip)
v1.3.10(Jul 24, 2021)
New Features

Register CrissCrossAttention into plugin layers (#1189)

Support custom hooks in IterBasedRunner (#1193)

Add ONNX export support to torch.roll (#1194)

Add flat cosine lr updater (#1066)

Add logger for initialization of parameters (#1150)

Improvements

Add is_tracing to wrap torch.jit.is_tracing in different versions of pytorch #1187

Revise the interface of upfirdn2d function (#1195)

Improve digit_version (#1185)

Add TERMINOLOGY.md (#895)

Bug Fixes

Fix potential integer overflow in imequalize (#1198)

Fix floor/ceil error in deform_attn for windows (#1037)

Resolves the duplicate logs issue (#1164)

Fix example (#1074)

Prevent kernel_normalizer to change mask dtype (#1210)

Notes

Since PyTorch 1.9 has been released, we plan to end the support for PyTorch 1.3 and 1.4 in the future releases. We will still provide pre-compiled packages of these two versions for three months, but the reliability and technical support are not guaranteed.

Contributors

A total of 19 developers contributed to this release.

@yeliudev, @007gzs, @wangg12, @qizhuli, @ddonatien, @mzr1996, @Junjun2016, @grimoire, @nbei, @dreamerlin, @jshilong, @kennymckormick, @zhouzaida, @ZwwWayne, @hellock, @MeowZheng, @innerlee, @xvjiarui, @RunningLeon
Source code(tar.gz)
Source code(zip)

OpenMMLab Computer Vision Foundation

Related tags

Overview

Introduction

Installation

FAQ

Citation

Contributing

License

Comments

Motivation

Modification

BC-breaking (Optional)

Use cases (Optional)

Checklist

Motivation

Modification

BC-breaking (Optional)

Use cases (Optional)

Checklist

Motivation

Modification

BC-breaking (Optional)

Use cases (Optional)

Checklist

Overview

Design

APIs

Initialization

before_run

before_train_xxx

after_train_xxx

Usages

Config

Migration

Modification for MM repos:

Releases(v1.7.1)

v1.7.1(Jan 3, 2023)

Features

Bug fixes

Documentations

Others

Contributors

New Contributors

v2.0.0rc3(Nov 25, 2022)

Features

Improvements

Bug fixes

Documentations

Others

Contributors

New Contributors

v1.7.0(Nov 3, 2022)

Highlights

Features

Improvements

Bug fixes

Documentations

Others

Contributors

New Contributors

v2.0.0rc2(Oct 26, 2022)

Features

Bug fixes

Documentations

Others

New Contributors

Contributors

v1.6.2(Sep 29, 2022)

Features

Improvements

Bug fixes

Documentations

Others

New Contributors

v2.0.0rc1(Aug 31, 2022)

Highlights

v1.6.1(Jul 26, 2022)

Features

Improvements