BriVL for Building Applications
This repo is used for illustrating how to build applications by using BriVL model.
This repo is re-implemented from following projects:
- Source Code of BriVL 1.0
- Model of BriVL 1.0
- Bottom Up Attention For Application
- bottom-up-attention
- bottom-up-attention.pytorch
Online Demo built by BriVL
Contents
This repo contains two parts:
- Bounding Box Extractor:
./bbox_extractor
- BriVL Feature Extractor:
./BriVL
Test this Pipeline
Test image has been saved in ./bbox_extractor/feature_extractor
, test with following command:
python3 main.py --brivl_cfg BriVL/cfg/BriVL_cfg.yml --brivl_weights BriVL/weights/brivl-weights.pth
Download Models
- bua-caffe-frcn-r101_with_attributes.pth ->
/bbox_extractor/weights
- chinese-roberta-wwm-ext ->
/BriVL/weights/hfl
- tf_efficientnet_b5_ns-6f26d0cf.pth ->
/BriVL/weights
- brivl-weights.pth* ->
/BriVL/weights
Requirements
- Python >= 3.6
- PyTorch >= 1.4
- Cuda >= 9.2 and cuDNN
- Detectron2 <= 0.3
- Transformers
Important: The version of Detectron2 should be 0.3 or below.
Install Pre-Built Detectron2 (Linux only)
Choose from this table to install v0.3 (Nov 2020):
CUDA | torch 1.7 | torch 1.6 | torch 1.5 |
---|---|---|---|
11.0 |
install
|
||
10.2 |
install
|
install
|
install
|
10.1 |
install
|
install
|
install
|
9.2 |
install
|
install
|
install
|
cpu |
install
|
install
|
install
|
More Resources
* indicates an application is needed.
Contact
This repo is maintained by Chuhao JIn(@jinchuhao).