dd6fe347创建于 4月9日历史提交

文件	最后提交记录	最后更新时间
config	init	4 年前
mmaction	!4685 [fix] 修改算子调用方式 * fix code clean all_py about Operator modification * fix code clean * fix op way first	2 年前
mmcv_need	!4685 [fix] 修改算子调用方式 * fix code clean all_py about Operator modification * fix code clean * fix op way first	2 年前
test	!4671 【fix】批量修改模型python版本，兼容环境上的python3.8版本 * fix python version	3 年前
README.md	!5805 Network address of models to be rectified: 21 Merge pull request !5805 from Yss/network_declaration_21	2 年前
public_address_statement.md	fix link validity Co-authored-by: frozenleaves<914814442@qq.com> # message auto-generated for no-merge-commit merge: !7517 merge master into master fix link validity Created-by: frozenn Commit-by: frozenleaves Merged-by: ascend-robot Description: ## Motivation Please describe the motivation of this PR and the goal you want to achieve through this PR. ## Modification Please briefly describe what modification is made in this PR. ## Self-test (Optional) If modifications to this PR may cause/fix function/accuracy/performance DTSs/issues, a self-inspection record needs to be attached. ## BC-breaking (Optional) If there are compatibility issues, such as dependencies on cann/torch_npu versions, they need to be explained in the PR. ## Checklist Before PR: - [ ] The new code needs to comply with the Clean Code specification. - [ ] The PR content is self-checked, and the expression can be clear and the writing standardized After PR: - [ ] CLA has been signed and all committers have signed the CLA in this PR. - [ ] The ci-pipeline is passed, Code Check is passed. See merge request: Ascend/ModelZoo-PyTorch!7517	1 个月前
requirements.txt	[众智][PyTorch]整改模型中的requirements.txt文件，删除torch,apex Signed-off-by: bailang <bailang12@h-partners.com>	3 年前
test.py	init	4 年前
train.py	init	4 年前

SlowFast

SlowFast 模型在 Kinetics400 数据集上的实现，主要修改自 open-mmlab/mmaction2 源码

环境准备

安装 Pytorch 和混合精度训练工具 Apex
安装依赖 pip install -r requirements.txt
下载 kinetics400 数据集

注：有关依赖中的 decord，若在 x86 平台上，可以直接通过 pip install decord 进行安装，若在 ARM Linux 平台上，则需通过源码编译安装，具体参见 dmlc/decord

安装 MMCV

下载 mmcv 源码

git clone -b v1.3.9 https://github.com/open-mmlab/mmcv.git
mv mmcv/ mmcv-master/
mv mmcv-master/mmcv ./
rm -rf mmcv-master/

替换 mmcv 中的文件

/bin/cp -f mmcv_need/base_runner.py mmcv/runner/base_runner.py
/bin/cp -f mmcv_need/builder.py mmcv/runner/optimizer/builder.py
/bin/cp -f mmcv_need/checkpoint.py mmcv/runner/hooks/checkpoint.py
/bin/cp -f mmcv_need/data_parallel.py mmcv/parallel/data_parallel.py
/bin/cp -f mmcv_need/dist_utils.py mmcv/runner/dist_utils.py
/bin/cp -f mmcv_need/distributed.py mmcv/parallel/distributed.py
/bin/cp -f mmcv_need/epoch_based_runner.py mmcv/runner/epoch_based_runner.py
/bin/cp -f mmcv_need/iter_timer.py mmcv/runner/hooks/iter_timer.py
/bin/cp -f mmcv_need/optimizer.py mmcv/runner/hooks/optimizer.py
/bin/cp -f mmcv_need/test.py mmcv/engine/test.py
/bin/cp -f mmcv_need/transformer.py mmcv/cnn/bricks/transformer.py

训练

训练阶段，脚本调用 train.py 进行训练

# 1p train perf
bash test/train_performance_1p.sh --data_path=/opt/npu

# 8p train perf
bash test/train_performance_8p.sh --data_path=/opt/npu

# 8p train full
bash test/train_full_8p.sh --data_path=/opt/npu

# 8p eval 
bash test/train_eval_8p.sh --data_path=/opt/npu

# finetuning
bash test/train_finetune_1p.sh --data_path=/opt/npu

注: 可以通过修改 --data_path 来指定数据集文件夹的位置，例如，你的数据集地址为：/opt/npu/kinetics400，可设置 --data_path=/opt/npu

Log Path:

slowfast_performance_1p.log # 1p 训练下性能测试日志
slowfast_performance_8p.log # 8p 训练下性能测试日志
slowfast_full_1p.log # 1p 完整训练下性能和精度测试日志
slowfast_full_8p.log # 8p 完整训练下性能和精度测试日志
slowfast_eval_8p.log # 8p 测试模型验证集精度日志
slowfast_finetune_1p.log # 1p 下 fine-tuning 日志

TSM-NonLocal 训练结果

top1 acc (单view)	FPS	Epochs	AMP_Type	Device
-	11	1	O1	1p Npu
33.43 (30 epochs)	81	256	O1	8p Npu
-	14	1	O1	1p Gpu
30.24 (30 epochs)	86	256	O1	8p Gpu

注：源仓库模型单 view 测试为 70.19 (256 epochs)，30 views 测试为 77.0（256 epochs）

公网地址说明

代码涉及公网地址参考 public_address_statement.md