RT-DETR源码学习bug记录

事情是这样的，我最近想学习RT-DETR的源码，那就开始吧！

1. 找到官网，找到pytorch版本。

https://github.com/lyuwenyu/RT-DETR/tree/main

2.只想下载一个子目录，方法：

https://download-directory.github.io/，然后在搜索框填入https://github.com/lyuwenyu/RT-DETR/tree/main/rtdetr_pytorch，回车即可。

3. 下载好之后，拖入服务器。亲测：创建好环境，激活，进入文件夹，然后pip requirements，即可运行

conda create -n rtdetr python=3.8
conda activate rtdetr
cd /home/cj/workspace/python/studyModel
git clone https://github.com/lyuwenyu/RT-DETR.git
cd /home/cj/workspace/python/studyModel/RT-DETR/rtdetr_pytorch
pip install -r requirements.txt

4.这里报错卡了2天，扫不到包，原来要设为根目录

5. 明明可以跑起来，但是今天打开有报错。

/home/cj/miniconda3/envs/rtdetr/bin/python tools/train.py -c configs/rtdetr/rtdetr_r50vd_6x_coco.yml 
Traceback (most recent call last):
  File "tools/train.py", line 9, in <module>
    import src.misc.dist as dist 
  File "/mnt/data_disk_1/cj/workspace/python/studyModel/RT-DETR/rtdetr_pytorch/tools/../src/__init__.py", line 2, in <module>
    from . import data 
  File "/mnt/data_disk_1/cj/workspace/python/studyModel/RT-DETR/rtdetr_pytorch/tools/../src/data/__init__.py", line 2, in <module>
    from .coco import *
  File "/mnt/data_disk_1/cj/workspace/python/studyModel/RT-DETR/rtdetr_pytorch/tools/../src/data/coco/__init__.py", line 1, in <module>
    from .coco_dataset import (
  File "/mnt/data_disk_1/cj/workspace/python/studyModel/RT-DETR/rtdetr_pytorch/tools/../src/data/coco/coco_dataset.py", line 14, in <module>
    from torchvision import datapoints
  File "/home/cj/miniconda3/envs/rtdetr/lib/python3.8/site-packages/torchvision/datapoints/__init__.py", line 1, in <module>
    from torchvision import _BETA_TRANSFORMS_WARNING, _WARN_ABOUT_BETA_TRANSFORMS
ImportError: cannot import name '_BETA_TRANSFORMS_WARNING' from 'torchvision' (/home/cj/miniconda3/envs/rtdetr/lib/python3.8/site-packages/torchvision/__init__.py)

Process finished with exit code 1

💬 建议
你可以执行以下命令，检查当前环境中 torch 和 torchvision 的版本：

python -c "import torch; import torchvision; print(torch.__version__, torchvision.__version__)"
2.3.0 0.18.1a0

你当前的环境版本如下：

torch: 2.3.0

torchvision: 0.18.1a0（注意这个是开发版，即 alpha 版本）

🚨 问题根源
你安装的 torchvision 0.18.1a0 是一个 开发预览版本（alpha），而这个版本中可能已经把 _BETA_TRANSFORMS_WARNING 移除了或重构了，因此导致了你训练脚本中导入失败的问题。

这类问题在 alpha 版本中很常见，不推荐用于生产或稳定训练流程。
回退到稳定版本（推荐）
回退到与 PyTorch 2.3 匹配的 torchvision 稳定版，比如 0.18.0：

pip install torchvision==0.18.0 --force-reinstall

成功，但是报不兼容

ERROR: pip's dependency resolver does not currently take into account all the packages that are installed. This behaviour is the source of the following dependency conflicts.
torchaudio 2.4.1 requires torch==2.4.1, but you have torch 2.3.0 which is incompatible.

不管，继续运行，然后报错：

/home/cj/miniconda3/envs/rtdetr/bin/python tools/train.py -c configs/rtdetr/rtdetr_r50vd_6x_coco.yml 
Traceback (most recent call last):
  File "tools/train.py", line 9, in <module>
    import src.misc.dist as dist 
  File "/mnt/data_disk_1/cj/workspace/python/studyModel/RT-DETR/rtdetr_pytorch/tools/../src/__init__.py", line 2, in <module>
    from . import data 
  File "/mnt/data_disk_1/cj/workspace/python/studyModel/RT-DETR/rtdetr_pytorch/tools/../src/data/__init__.py", line 2, in <module>
    from .coco import *
  File "/mnt/data_disk_1/cj/workspace/python/studyModel/RT-DETR/rtdetr_pytorch/tools/../src/data/coco/__init__.py", line 1, in <module>
    from .coco_dataset import (
  File "/mnt/data_disk_1/cj/workspace/python/studyModel/RT-DETR/rtdetr_pytorch/tools/../src/data/coco/coco_dataset.py", line 14, in <module>
    from torchvision import datapoints
ImportError: cannot import name 'datapoints' from 'torchvision' (/home/cj/miniconda3/envs/rtdetr/lib/python3.8/site-packages/torchvision/__init__.py)

Process finished with exit code 1

重新装兼容版吧

pip uninstall torch torchvision torchaudio -y
pip install torch==2.4.1 torchvision==0.19.1 torchaudio==2.4.1 --index-url https://download.pytorch.org/whl/cu118

没成功，回退版本

pip install torchvision==0.18.0 --force-reinstall

（1）在PyCharm中，Structure是一个非常有用的功能，它可以帮助开发者快速浏览和理解当前文件的代码结构：alt+7