Skip to content

This issue was moved to a discussion.

You can continue the conversation there. Go to discussion →

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

【模型压缩推全计划】为六大套件新增模型压缩功能 #10657

Closed
shiyutang opened this issue Aug 16, 2023 · 8 comments
Closed
Assignees
Labels
Code PR is needed This issue could inspire a code PR good first issue Good for newcomers

Comments

@shiyutang
Copy link
Collaborator

shiyutang commented Aug 16, 2023

背景

目前各套件的模型压缩能力参差不齐,而模型压缩作为部署之前的一步,可以在不损害或者少量损害模型精度的情况下,对模型的能耗,速度、大小都有显著的改善。因此为了对各套件的模型压缩进行推全,我们提出了基于PaddleSlim的ACT为各大套件新增模型压缩功能的计划。

解决步骤:

  1. 查看Paddleslim的ACT文档,学习ACT,熟悉基于不同任务进行ACT中量化压缩的主要步骤。

  2. 为某一任务接入ACT中的量化训练压缩功能:在适配paddle2.5,paddleslim2.5和套件develop版本的前提下,仿照pp-Liteseg接入ACT的PR,在各大套件的部署文件夹下(例如分割为PaddleSeg/deploy/slim/act) 接入ACT。下面以分割为例说明需要包括的新增文件内容:

    • 增加ACT启动脚本 run_seg.py 在 PaddleSeg/deploy/slim/act/run_seg.py
    • 增加量化训练配置文件在 PaddleSeg/deploy/slim/act/configs/ppliteseg_qat.py
    • 增加用于验证精度和速度的推理脚本在PaddleSeg/deploy/slim/act/test_seg.py
  3. 参考paddleslim的ACT使用文档,对量化训练模型进行导出,并验证不同模型在量化蒸馏训练前后的精度和速度对比,验证通过后,撰写量化训练使用文档,增加相关Readme在 PaddleSeg/deploy/slim/act/readme.md。

  4. 提交PR到对应模型的仓库中。

注:不同套件的部署目录如下,将上述分割的目录进行相应替换即可:

  • PaddleClas: PaddleClas/deploy/slim/act
  • PaddleDetection: PaddleDetection/deploy/auto_compression(该目录下的文件为之前接入的文件,可以复用)
  • Paddle3D:Paddle3D/deploy/slim_act
  • PaddleOCR: PaddleOCR/deploy/slim/act
  • PaddleSeg: PaddleSeg/deploy/slim/act/

模型列表

同一个任务仅有配置文件差异,一个任务需要完成对应的模型配置,领取通用目标检测任务的开发者将认领四个检测模型。

任务名称 模型参数名称 报名开发者
通用目标检测 DINO_r50_4scale @MINGtoMING
通用目标检测 RT-DETR-L @MINGtoMING
通用目标检测 PP-YOLOE+_crn_l_80e @MINGtoMING
通用目标检测 PP-PicoDet_s_320_LCNet @MINGtoMING
通用图像分类 SwinTransformer_base_patch4_window7_224 @vax521
通用图像分类 PP-LCNet_x1_0 @vax521
通用图像分类 CLIP_vit_base_patch16_224 @vax521
通用图像分类 PP-HGNet_small @vax521
通用图像分类 ResNet50 @vax521
通用图像分类 MobileNetV3_small_x1_0 @vax521
通用文本检测 PP-OCRv4-server-det @xu-peng-7
通用文本检测 PP-OCRv4-mobile-det @xu-peng-7
通用文本识别 PP-OCRv4-server-rec
通用文本识别 PP-OCRv4-mobile-rec
通用图像分割 OCRNet_HRNetW48 @marshall-dteach
通用图像分割 SegFormer-B0 @marshall-dteach
通用图像分割 PP-LiteSeg-Tiny @marshall-dteach
通用图像分割 PP-MobileSeg-Base @marshall-dteach
3D目标检测 CenterPoint_pillars_02voxel
3D目标检测 PETRv2_v99
3D目标检测 PETRv1_v99
3D目标检测 CaDDN_ocrnet_hrnet_w18
通用主体检测 PP-ShiTuv2-det @Zheng-Bicheng
通用检索特征 PP-ShiTuv2-rec @Zheng-Bicheng
通用版面分析 PP-Structurev2-layout
通用表格识别 PP-Structurev2-SLANet
通用关键信息抽取 PP-Structurev2-vi-layoutxlm
@WilliamQf-AI
Copy link
Contributor

WilliamQf-AI commented Aug 16, 2023

队伍名:风清扬
队伍成员微信昵称:WilliamQf-AI
功能描述:通用版面分析

@shiyutang shiyutang added the Code PR is needed This issue could inspire a code PR label Aug 16, 2023
@WilliamQf-AI
Copy link
Contributor

队伍名:风清扬
队伍成员微信昵称:WilliamQf-AI
功能描述:通用表格识别

@ToddBear
Copy link
Collaborator

队伍名:飞桨小队
队伍成员微信昵称:zejia_su
功能描述:通用文本识别

@marshall-dteach
Copy link

队伍名:德布罗意波
队伍成员微信昵称:TB1250859983
功能描述:通用图像分割

@xu-peng-7
Copy link
Contributor

队伍名:Dec20b
队伍成员微信昵称:muted
功能描述:通用文本检测

@vax521
Copy link

vax521 commented Aug 30, 2023

队伍名:晓飞队
队伍成员微信昵称:晓飞
功能描述:通用图像分类

@Zheng-Bicheng
Copy link

队伍名:Zheng-Bicheng
队伍成员微信昵称:Zheng-Bicheng
功能描述:通用主体检测

@Zheng-Bicheng
Copy link

队伍名:Zheng-Bicheng
队伍成员微信昵称:Zheng-Bicheng
功能描述:通用检索特征

@PaddlePaddle PaddlePaddle locked and limited conversation to collaborators Jun 5, 2024
@SWHL SWHL converted this issue into discussion #12736 Jun 5, 2024

This issue was moved to a discussion.

You can continue the conversation there. Go to discussion →

Labels
Code PR is needed This issue could inspire a code PR good first issue Good for newcomers
Projects
None yet
Development

No branches or pull requests

8 participants