Rotating frame target detection mmrotate v0.3.1 learning model

1、 We basically divide the model components into 5 Types

1、backbone: usually an FCN network to extract feature maps, e.g., ResNet, Swin.

2、neck: the component between backbones and heads, e.g., FPN, ReFPN.

3、head: the component for specific tasks, e.g., bbox prediction.

4、roi extractor: the part for extracting RoI features from feature maps, e.g., RoI Align Rotated.

5、loss: the component in head for calculating losses, e.g., FocalLoss, GWDLoss, and KFIoULoss.

2、 Develop new components

2.1 Add a new backbone

Here we have MobileNet As an example to show how to develop new components

1. Define a new backbone (e.g. MobileNet)

Create a new file mmrotate/models/backbones/mobilenet.py.

import torch.nn as nn

from mmrotate.models.builder import ROTATED_BACKBONES


@ROTATED_BACKBONES.register_module()
class MobileNet(nn.Module):

    def __init__(self, arg1, arg2):
        pass

    def forward(self, x):  # should return a tuple
        pass

2.2 The import module

You can either add the following line to mmrotate/models/backbones/init.py

from .mobilenet import MobileNet

or alternatively add

custom_imports = dict(
    imports=['mmrotate.models.backbones.mobilenet'],
    allow_failed_imports=False)

To the configuration file , To avoid modifying the original code .

2.3. Use the backbone in your config file

model = dict(
    ...
    backbone=dict(
        type='MobileNet',
        arg1=xxx,
        arg2=xxx),
    ...