TinaFace搭建环境训练Wider face数据集

一直遇到的问题是，我重装咯3次虚拟环境，在window和ubantu 系统都遇到这个问题：

 File "C:\Users\HP\AppData\Local\Temp\pip-build-env-_u60y33_\overlay\Lib\site-packages\setuptools\build_meta.py", line 338, in run_setup
      exec(code, locals())
    File "<string>", line 5, in <module>
  ModuleNotFoundError: No module named 'torch'
  error: subprocess-exited-with-error

  × Getting requirements to build editable did not run successfully.
  │ exit code: 1
  ╰─> See above for output.

  note: This error originates from a subprocess, and is likely not a problem with pip.
  full command: 'D:\anaconda3\envs\vedadet\python.exe' 'D:\anaconda3\envs\vedadet\lib\site-packages\pip\_vendor\pyproject_hooks\_in_process\_in_process.py' get_requires_for_build_editable 'C:\Users\HP\AppData\Local\Temp\tmpyge6zm3k'
  cwd: D:\Vir_env\vedadet-main
  Getting requirements to build editable ... error
error: subprocess-exited-with-error

× Getting requirements to build editable did not run successfully.
│ exit code: 1
╰─> See above for output.

note: This error originates from a subprocess, and is likely not a problem with pip.

 cwd: C:\Users\HP\AppData\Local\Temp\pip-install-mwuj0gs3\mmpycocotools_e4f4efa3382448a99d5d82096e7a9c2a\
  Building wheel for mmpycocotools (setup.py) ... error
  ERROR: Failed building wheel for mmpycocotools
  Running setup.py clean for mmpycocotools
  Running command python setup.py clean
  running clean
  removing 'build\temp.win-amd64-cpython-38' (and everything under it)
  removing 'build\lib.win-amd64-cpython-38' (and everything under it)
  'build\bdist.win-amd64' does not exist -- can't clean it
  'build\scripts-3.8' does not exist -- can't clean it
  removing 'build'
Failed to build mmpycocotools
ERROR: Could not build wheels for mmpycocotools, which is required to install pyproject.toml-based projects

尝试咯网上很多的解决方式，都不行。然后我就不按照官网说的安装环境，直接就是按照mmdetetion的环境安装环境，最后环境是装好咯。

---

接下来时安装环境的步骤：

①下载安装包：

官网的网址：https://github.com/Media-Smart/vedadet/tree/main/configs/trainval/tinaface

在官网下载安装包：Media-Smart/vedadet: A single stage object detection toolbox based on PyTorch (github.com)

---

 

---

 ②下载虚拟环境

 然后打开mmdetection的官网按照这个步骤下载：GET STARTED — MMDetection 3.1.0 documentation

记得在下载cuda和pytorch的时候选择10.2，官网是在1.6版本测试的，反正cuda版本一定要选择10.2，因为我在后面时候发现莫名其妙会说你的cuda版本是10.2，就算你的不是。如果不信你可以下其他的，回来给我反馈是不是我的问题。这个过程可能还是遇到pytorch版本不适配，多下几次就行哈哈。

旧版本的pytorch下载网址：Previous PyTorch Versions | PyTorch

我下载cuda10.2,pytorch1.6是可以的，但是mmdetection要求pytorch1.8以上，所以我尝试下载1.8的，但是遇到我的cuda 不存在这个情况，我人麻咯：

pip install -U openmim
mim install mmengine
mim install "mmcv>=2.0.0"

再按照tinafce官网下载vedadet: Media-Smart/vedadet: A single stage object detection toolbox based on PyTorch (github.com)

---

---

③下载数据集和权重文件：

按照这个步骤把数据放在对应的位置（一定要严格按照这个来操作）

 在这个网址下载训练权重：tinaface - Google 云端硬盘

在这个网址下载数据集：wider_face |TensorFlow Datasets

在这个网址下载数据集的annotation:GitHub - sovrasov/wider-face-pascal-voc-annotations: WIDER FACE annotations converted to the Pascal VOC XML format

将WIDER_train_annotations 和 WIDER_val_annotations文件名更改为Annotations，并分别放置在WIDER_train 和 WIDER_val两个文件下，并将改名前WIDER_train_annotations 和 WIDER_val_annotations两个文件夹中的train.txt和val.txt分别移动至data/WIDERFace/WIDER_train和WIDER_val文件下。（反正你在代码里面写的路径要和你放的路径一样）

记着这里有一个坑：就是在train.txt和val.txt文档文件上面的最后一行分别有train和val这2行，要删掉不然会报错说没有train.xml这个文件的。

 

---

---

④训练和测试

测试：我用的是tinaface做训练和测试：（不是retinanet下面这2指令都可以用来测试）

CUDA_VISIBLE_DEVICES="0" python tools/infer.py configs/infer/tinaface/tinaface_r50_fpn_bn.py picture/face.jpg（换成你的图片路径）

CUDA_VISIBLE_DEVICES="0" python tools/infer.py configs/infer/tinaface/tinaface_r50_fpn_gn_dcn.py picture/face.jpg（换成你的图片路径）

 

 

训练：我用的是tinaface做训练和测试：（不是retinanet)这2个都可以用来训练）

CUDA_VISIBLE_DEVICES="0" python tools/trainval.py configs/trainval/tinaface/tinaface_r50_fpn_bn.py

CUDA_VISIBLE_DEVICES="0" python tools/trainval.py configs/trainval/tinaface/tinaface_r50_fpn_gn_dcn.py

 参考链接：

 TinaFace完整复现及其搭建自己的网络模型_JaciusCV的博客-CSDN博客

mmdetection在wider face数据集上训练人脸检测网络_JaciusCV的博客-CSDN博客