docker of mmdetection3 not support CUDA

docker image [heartexlabs/label-studio-ml-backend:mmdetection3-master](https://hub.docker.com/layers/heartexlabs/label-studio-ml-backend/mmdetection3-master/images/sha256-30bc39ec2e3d45f2fc00d8345dad501a8a8ae37bd2fe76f6a2c5619d126b4bbd) encountered an issue.

```
RuntimeError( RuntimeError: Cannot re-initialize CUDA in forked subprocess. To use CUDA with multiprocessing, you must use the 'spawn' start method
```

same as
https://github.com/HumanSignal/label-studio-ml-backend/issues/560
https://github.com/HumanSignal/label-studio-ml-backend/issues/79

Overriding the command to remove the `--preload` parameter will allow it to run normally.

```yaml
services:
  mmdetection3:
    image: heartexlabs/label-studio-ml-backend:mmdetection3-master
    container_name: mmdetection3
    # https://github.com/HumanSignal/label-studio-ml-backend/issues/79
    command: gunicorn --bind :9090 --workers 1 --threads 8 --timeout 0 _wsgi:app
    env_file: external_env-mmdetection3.env
    ports:
      - '9090'
    networks:
      - labelstudio
    volumes:
      - data:/data
```

after that

```sh
[2025-01-08 06:42:00,804] [ERROR] [label_studio_ml.exceptions::exception_f::53] Traceback (most recent call last):
File "/opt/conda/lib/python3.9/site-packages/label_studio_ml/exceptions.py", line 39, in exception_f
return f(*args, **kwargs)
File "/opt/conda/lib/python3.9/site-packages/label_studio_ml/api.py", line 69, in _predict
response = model.predict(tasks, context=context, **params)
File "/app/mmdetection.py", line 160, in predict
prediction = self.predict_one_task(task)
File "/app/mmdetection.py", line 167, in predict_one_task
model_results = inference_detector(model, image_path).pred_instances
File "/opt/conda/lib/python3.9/site-packages/mmdet/apis/inference.py", line 189, in inference_detector
results = model.test_step(data_)[0]
File "/opt/conda/lib/python3.9/site-packages/mmengine/model/base_model/base_model.py", line 145, in test_step
return self._run_forward(data, mode='predict') # type: ignore
File "/opt/conda/lib/python3.9/site-packages/mmengine/model/base_model/base_model.py", line 361, in _run_forward
results = self(**data, mode=mode)
File "/opt/conda/lib/python3.9/site-packages/torch/nn/modules/module.py", line 1190, in _call_impl
return forward_call(*input, **kwargs)
File "/opt/conda/lib/python3.9/site-packages/mmdet/models/detectors/base.py", line 94, in forward
return self.predict(inputs, data_samples)
File "/opt/conda/lib/python3.9/site-packages/mmdet/models/detectors/single_stage.py", line 110, in predict
results_list = self.bbox_head.predict(
File "/opt/conda/lib/python3.9/site-packages/mmdet/models/dense_heads/base_dense_head.py", line 197, in predict
predictions = self.predict_by_feat(
File "/opt/conda/lib/python3.9/site-packages/mmdet/models/dense_heads/yolo_head.py", line 280, in predict_by_feat
results = self._bbox_post_process(
File "/opt/conda/lib/python3.9/site-packages/mmdet/models/dense_heads/base_dense_head.py", line 485, in _bbox_post_process
det_bboxes, keep_idxs = batched_nms(bboxes, results.scores,
File "/opt/conda/lib/python3.9/site-packages/mmcv/ops/nms.py", line 303, in batched_nms
dets, keep = nms_op(boxes_for_nms, scores, **nms_cfg_)
File "/opt/conda/lib/python3.9/site-packages/mmengine/utils/misc.py", line 395, in new_func
output = old_func(*args, **kwargs)
File "/opt/conda/lib/python3.9/site-packages/mmcv/ops/nms.py", line 127, in nms
inds = NMSop.apply(boxes, scores, iou_threshold, offset, score_threshold,
File "/opt/conda/lib/python3.9/site-packages/mmcv/ops/nms.py", line 27, in forward
inds = ext_module.nms(
RuntimeError: nms_impl: implementation for device cuda:0 not found.
```

Same as https://github.com/open-mmlab/mmdetection/issues/6765

Reinstalling `mmcv` in the container and restarting it resolves the issue.

```sh
root@f0ea7e29bc01:/app# mim uninstall mmcv
Found existing installation: mmcv 2.1.0
Uninstalling mmcv-2.1.0:
Would remove:
/opt/conda/lib/python3.9/site-packages/mmcv-2.1.0.dist-info/*
/opt/conda/lib/python3.9/site-packages/mmcv/*
Proceed (Y/n)? y
Successfully uninstalled mmcv-2.1.0
WARNING: Running pip as the 'root' user can result in broken permissions and conflicting behaviour with the system package manager, possibly rendering your system unusable.It is recommended to use a virtual environment instead: https://pip.pypa.io/warnings/venv. Use the --root-user-action option if you know what you are doing and want to suppress this warning.
root@f0ea7e29bc01:/app# mim install mmcv==2.1.0
Looking in links: https://download.openmmlab.com/mmcv/dist/cu116/torch1.13.0/index.html
WARNING: Retrying (Retry(total=4, connect=None, read=None, redirect=None, status=None)) after connection broken by 'NewConnectionError(': Failed to establish a new connection: [Errno -2] Name or service not known')': /mmcv/dist/cu116/torch1.13.0/index.html
Collecting mmcv==2.1.0
Downloading https://download.openmmlab.com/mmcv/dist/cu116/torch1.13.0/mmcv-2.1.0-cp39-cp39-manylinux1_x86_64.whl (97.6 MB)
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 97.6/97.6 MB 427.3 kB/s eta 0:00:00
Requirement already satisfied: addict in /opt/conda/lib/python3.9/site-packages (from mmcv==2.1.0) (2.4.0)
Requirement already satisfied: mmengine>=0.3.0 in /opt/conda/lib/python3.9/site-packages (from mmcv==2.1.0) (0.10.3)
Requirement already satisfied: numpy in /opt/conda/lib/python3.9/site-packages (from mmcv==2.1.0) (1.26.4)
Requirement already satisfied: packaging in /opt/conda/lib/python3.9/site-packages (from mmcv==2.1.0) (24.2)
Requirement already satisfied: Pillow in /opt/conda/lib/python3.9/site-packages (from mmcv==2.1.0) (10.4.0)
Requirement already satisfied: pyyaml in /opt/conda/lib/python3.9/site-packages (from mmcv==2.1.0) (6.0.2)
Requirement already satisfied: yapf in /opt/conda/lib/python3.9/site-packages (from mmcv==2.1.0) (0.43.0)
Requirement already satisfied: opencv-python>=3 in /opt/conda/lib/python3.9/site-packages (from mmcv==2.1.0) (4.10.0.84)
Requirement already satisfied: matplotlib in /opt/conda/lib/python3.9/site-packages (from mmengine>=0.3.0->mmcv==2.1.0) (3.9.4)
Requirement already satisfied: rich in /opt/conda/lib/python3.9/site-packages (from mmengine>=0.3.0->mmcv==2.1.0) (13.4.2)
Requirement already satisfied: termcolor in /opt/conda/lib/python3.9/site-packages (from mmengine>=0.3.0->mmcv==2.1.0) (2.5.0)
Requirement already satisfied: platformdirs>=3.5.1 in /opt/conda/lib/python3.9/site-packages (from yapf->mmcv==2.1.0) (4.3.6)
Requirement already satisfied: tomli>=2.0.1 in /opt/conda/lib/python3.9/site-packages (from yapf->mmcv==2.1.0) (2.2.1)
Requirement already satisfied: contourpy>=1.0.1 in /opt/conda/lib/python3.9/site-packages (from matplotlib->mmengine>=0.3.0->mmcv==2.1.0) (1.3.0)
Requirement already satisfied: cycler>=0.10 in /opt/conda/lib/python3.9/site-packages (from matplotlib->mmengine>=0.3.0->mmcv==2.1.0) (0.12.1)
Requirement already satisfied: fonttools>=4.22.0 in /opt/conda/lib/python3.9/site-packages (from matplotlib->mmengine>=0.3.0->mmcv==2.1.0) (4.55.3)
Requirement already satisfied: kiwisolver>=1.3.1 in /opt/conda/lib/python3.9/site-packages (from matplotlib->mmengine>=0.3.0->mmcv==2.1.0) (1.4.7)
Requirement already satisfied: pyparsing>=2.3.1 in /opt/conda/lib/python3.9/site-packages (from matplotlib->mmengine>=0.3.0->mmcv==2.1.0) (3.2.0)
Requirement already satisfied: python-dateutil>=2.7 in /opt/conda/lib/python3.9/site-packages (from matplotlib->mmengine>=0.3.0->mmcv==2.1.0) (2.9.0.post0)
Requirement already satisfied: importlib-resources>=3.2.0 in /opt/conda/lib/python3.9/site-packages (from matplotlib->mmengine>=0.3.0->mmcv==2.1.0) (6.4.5)
Requirement already satisfied: markdown-it-py>=2.2.0 in /opt/conda/lib/python3.9/site-packages (from rich->mmengine>=0.3.0->mmcv==2.1.0) (3.0.0)
Requirement already satisfied: pygments<3.0.0,>=2.13.0 in /opt/conda/lib/python3.9/site-packages (from rich->mmengine>=0.3.0->mmcv==2.1.0) (2.18.0)
Requirement already satisfied: zipp>=3.1.0 in /opt/conda/lib/python3.9/site-packages (from importlib-resources>=3.2.0->matplotlib->mmengine>=0.3.0->mmcv==2.1.0) (3.21.0)
Requirement already satisfied: mdurl~=0.1 in /opt/conda/lib/python3.9/site-packages (from markdown-it-py>=2.2.0->rich->mmengine>=0.3.0->mmcv==2.1.0) (0.1.2)
Requirement already satisfied: six>=1.5 in /opt/conda/lib/python3.9/site-packages (from python-dateutil>=2.7->matplotlib->mmengine>=0.3.0->mmcv==2.1.0) (1.16.0)
Installing collected packages: mmcv
Successfully installed mmcv-2.1.0
```

Referring to https://github.com/open-mmlab/mmdetection/issues/6765#issuecomment-2563953554 can resolve the issue.

change dockerfile like that

```dockerfile
ENV PYTHONUNBUFFERED=1 \
    PYTHONDONTWRITEBYTECODE=1 \
    PORT=${PORT:-9090} \
    PIP_CACHE_DIR=/.cache \
    WORKERS=1 \
    THREADS=8 \
    CUDA_HOME=/usr/local/cuda
ENV PATH="${CUDA_HOME}/bin:${PATH}"
ENV TORCH_CUDA_ARCH_LIST="6.0;6.1;7.0;7.5;8.0;8.6+PTX;8.9;9.0"
```

Start with the mmdetection3 Dockerfile and the gunicorn command described in the issue, then inspect mmdetection.py alongside the CUDA and mmcv traceback. Verify the image's CUDA architecture and worker configuration, and confirm that the rebuilt container performs inference without the forked-CUDA or missing CUDA NMS errors.

Written by the indexing model from the issue text.