| 文件 | 最后提交记录 | 最后更新时间 |
|---|---|---|
| 4 年前 | ||
| 4 年前 | ||
| 4 年前 | ||
| 4 年前 | ||
| 2 年前 | ||
| 2 年前 | ||
| 4 年前 | ||
| 3 年前 | ||
| 3 年前 | ||
| 4 年前 | ||
| 3 年前 | ||
| 4 年前 | ||
| 4 年前 | ||
| 2 年前 | ||
| 4 年前 | ||
| 4 年前 | ||
| 4 年前 | ||
| 8 个月前 | ||
| 4 年前 | ||
| 4 年前 | ||
| 4 年前 | ||
| 4 年前 |
GFocalV2
This implements training of GFocalV2 on the Coco dataset, mainly modified from pytorch/examples.
GFocalV2 Detail
As of the current date, Ascend-Pytorch is still inefficient for contiguous operations. Therefore, GFocalV2 model need to be modified in the following aspects:
- Converting tensors with the dynamic shapes into tensors with fixed shapes. (This is the hardest one)
- Several operations, like the sum of
INT64, are not supported on the NPU, so we modified tensors'dtypewhen needed - Framework bottlenecks lead to poor performance, so we improve the original code to improve the performance of the model
- We used Apex for mmdtection due to the hardware defects of the NPU
- ...
Requirements
- NPU配套的run包安装
- Python 3.7.5
- PyTorch(NPU版本)
- apex(NPU版本)
- MMCV v1.2.7
Document and data preparation
- 下载压缩modelzoo\contrib\PyTorch\cv\image_object_detection\GFocalV2 文件夹
- 于npu服务器解压GFocalV2压缩包
- 下载coco2017数据集
- 将coco数据集放于GFocalV2/data目录下，目录结构如下：
GFocalV2
├── configs
├── data
│ ├── coco
│ ├── annotations 796M
│ ├── train2017 19G
│ ├── val2017 788M
Download and modify mmcv
- 在GFocalV2目录上级目录，下载mmcv，最好是1.2.7版本的（版本要求是1.2.5以上，1.3.0以下）
git clone -b v1.2.7 --depth=1 https://github.com/open-mmlab/mmcv.git
- 进入GFocalV2目录，用mmcv_need里的文件替换mmcv中对应的文件
cp -f ./mmcv_need/_functions.py ./mmcv/mmcv/parallel/
cp -f ./mmcv_need/builder.py ./mmcv/mmcv/runner/optimizer/
cp -f ./mmcv_need/distributed.py ./mmcv/mmcv/parallel/
cp -f ./mmcv_need/data_parallel.py ./mmcv/mmcv/parallel/
cp -f ./mmcv_need/dist_utils.py ./mmcv/mmcv/runner/
cp -f ./mmcv_need/optimizer.py ./mmcv/mmcv/runner/hooks/
- 以下三个文件的替换是为了在log中打印出FPS的信息
cp -f ./mmcv_need/iter_timer.py ../mmcv/mmcv/runner/hooks/
cp -f ./mmcv_need/base_runner.py ../mmcv/mmcv/runner/
cp -f ./mmcv_need/epoch_based_runner.py ../mmcv/mmcv/runner/
Configure the environment
先source环境变量
source GFocalV2/test/env_npu.sh
- 配置安装mmcv
cd mmcv
MMCV_WITH_OPS=1 pip3.7 install -e .
cd ..
pip3 list | grep mmcv # 查看版本和路径
- 配置安装mmdet
cd GFocalV2
pip3 install -r requirements/build.txt
python3 setup.py develop
cd ..
- 修改apex中的113行，主要是为了支持O1，参考路径root/archiconda3/envs/fcos/lib/python3.7/site-packages/apex/amp/utils.py
if cached_x.grad_fn.next_functions[1][0].variable is not x:
改成
if cached_x.grad_fn.next_functions[0][0].variable is not x:
Train MODEL
进入GFocalV2目录下
1p
导入环境变量，修改train_full_1p.sh权限并运行
chmod +x ./test/train_full_1p.sh
bash ./test/train_full_1p.sh --data_path=./data/coco
8p
导入环境变量，修改train_full_8p.sh权限并运行
bash ./test/train_full_8p.sh --data_path=./data/coco
Eval
修改train_eval_1p.sh权限并运行
chmod +x ./test/train_eval_1p.sh
bash ./test/train_eval_1p.sh --data_path=./data/coco
finetuning
修改train_finetune_1p.sh权限并运行
chmod +x ./test/train_eval_1p.sh
bash ./test/train_finetune_1p.sh --data_path=./data/coco --checkpoint=xxx（可选，gfocal模型的权重文件）
Demo
source ./test/env_npu.sh
python3 demo.py --checkpoint xxx(可选，gfocal模型的权重文件，默认./work_dirs/gfocal_r50_fpn_1x/latest.pth） --img xxx(可选，测试图片）
GFocalV2 training result
| Acc@1 | FPS | Npu/Gpu_nums | Epochs | AMP_Type | Loss_Scale |
|---|---|---|---|---|---|
| 23.2 | 14.6 | 1p Gpu | 1 | O1 | 128.0 |
| 41.0 | 71.3 | 8p Gpu | 12 | O1 | 128.0 |
| 23.5 | 3.46 | 1p Npu | 1 | O1 | 128.0 |
| 40.9 | 27.15 | 8p Npu | 12 | O1 | 128.0 |
公网地址说明
代码涉及公网地址参考 public_address_statement.md