GCNet
This implements training of GCNet on the Coco dataset, mainly modified from pytorch/examples.
GCNet Detail
GCNet is initially described in arxiv. Via absorbing advantages of Non-Local Networks (NLNet) and Squeeze-Excitation Networks (SENet), GCNet provides a simple, fast and effective approach for global context modeling, which generally outperforms both NLNet and SENet on major benchmarks for various recognition tasks.
Requirements
- NPU配套的run包安装
- Python 3.7.5
- PyTorch(NPU版本)
- apex(NPU版本)
Document and data preparation
- 下载压缩GCNet文件夹
- 于npu服务器解压GCNet压缩包
- 准备coco数据集并放置在指定位置
Download and modify mmcv
- 下载mmcv-full，使用的版本为1.3.8; 下载mmdetection，使用的版本为1.2.7
git clone -b v1.3.8 https://github.com/open-mmlab/mmcv.git
git clone -b v2.10.0 https://github.com/open-mmlab/mmdetection.git
- 用GCNet/dependency/mmcv目录替换clone文件夹里mmcv的mmcv（mmcv/mmcv）
cp -f dependency/mmcv/_functions.py ./mmcv/mmcv/parallel/
cp -f dependency/mmcv/base_runner.py ./mmcv/mmcv/runner/
cp -f dependency/mmcv/builder.py ./mmcv/mmcv/runner/optimizer/
cp -f dependency/mmcv/checkpoint.py ./mmcv/mmcv/runner/
cp -f dependency/mmcv/context_block.py ./mmcv/mmcv/cnn/bricks
cp -f dependency/mmcv/data_parallel.py ./mmcv/mmcv/parallel/
cp -f dependency/mmcv/dist_utils.py ./mmcv/mmcv/runner/
cp -f dependency/mmcv/distributed.py ./mmcv/mmcv/parallel/
cp -f dependency/mmcv/epoch_based_runner.py ./mmcv/mmcv/runner/
cp -f dependency/mmcv/iter_based_runner.py ./mmcv/mmcv/runner/
cp -f dependency/mmcv/iter_timer.py ./mmcv/mmcv/runner/hooks/
cp -f dependency/mmcv/optimizer.py ./mmcv/mmcv/runner/hooks/
cp -f dependency/mmcv/roi_align.py ./mmcv/mmcv/ops/
或是pip安装mmcv-full后手动替换库文件
- 用GCNet/dependency/mmdet目录替换clone文件夹里mmdetection的mmdet（mmdetection/mmdet）
rm -r mmdetection/mmdet
cp -r dependency/mmdet mmdetection/
或是pip安装mmdet后手动替换库文件
Configure the environment
- 推荐使用conda管理
conda create -n gcnet --clone env # 复制一个已经包含依赖包的环境
conda activate gcnet
- 配置安装mmcv
source ./test/env_npu.sh
cd mmcv
export MMCV_WITH_OPS=1
export MAX_JOBS=8
python3 setup.py build_ext
python3 setup.py develop
pip3 list | grep mmcv # 查看版本和路径
- 配置安装mmdet
cd mmdetection
pip install -r requirements/build.txt
pip install -v -e . # or "python setup.py develop"
注意：
如果复制的conda环境安装过pycocotools包，需要卸载后重新安装mmpycocotools
pip uninstall pycocotools
pip install mmpycocotools
Train MODEL
进入GCNet文件夹下
cd GCNet
# 1p train_1p
bash ./test/train_full_1p.sh --data_path=数据集路径
# 8p train_8p
bash ./test/train_full_8p.sh --data_path=数据集路径
# 8p perf_8p
bash ./test/train_performance_8p.sh --data_path=数据集路径
# 1p perf_1p
bash ./test/train_performance_1p.sh --data_path=数据集路径
# 1p eval
bash ./test/eval.sh --weight_path=数据集路径
参考精度/性能
| 名称 | 精度(mAP) | 性能(fps) |
|---|---|---|
| GPU-1p | - | 8.47 |
| GPU-8p | 39.9 | 44.62 |
| NPU-1p | - | 0.52 |
| NPU-8p | 39.1 | 2.35 |
公网地址说明
代码涉及公网地址参考 public_address_statement.md