ImportError: Failed to import projects.<MODEL_NAME>.<model>

[charisoudis]

I cannot instantiate a custom model, Co-DETR given the weights.

After searching, it seems that there is a package named "projects" which prevents python from searching into the mmdetection directory (it is in the PYTHONPATH).

>>> from mmdet.apis import DetInferencer
>>> DetInferencer(weights='/workspace/out/checkpoints/co-detr/co_dino_5scale_r50_lsj_8xb2_1x_coco-69a72d67.pth')
>>> ...
>>> ImportError: Failed to import custom modules from {'allow_failed_imports': False, 'imports': ['projects.CO-DETR.codetr']}, the current sys.path is: 
    
    /mmdetection
    /workspace
    /opt/conda/lib/python310.zip
    /opt/conda/lib/python3.10
    /opt/conda/lib/python3.10/lib-dynload
    /opt/conda/lib/python3.10/site-packages
    /opt/conda/lib/python3.10/site-packages/smplx-0.1.28-py3.10.egg
    /tmp/tmp1_v1g0zy
You should set `PYTHONPATH` to make `sys.path` include the directory which contains your custom module
>>>
>>>
>>> import projects
>>> projects.__file__
'/opt/conda/lib/python3.10/site-packages/projects/__init__.py'
>>> 

My env is a docker image (thanosch/mmtorch:2.1.0-cuda121) with Python 3.10 and the latest version of mm:

• mmcv 2.1.0
• mmdet 3.2.0
• mmengine 0.9.0
• mmpose 1.2.0

[hhaAndroid]

@achariso This operation is not supported under the project. You need to prepare the configuration and weights.

[Jordan-Pierce]

@hhaAndroid can you please elaborate?

[LokeshJatangi]

@Jordan-Pierce I could solve the above error , by adding path to mmdetection to python path or through sys.path.append(path_to_mmedetection_dir)

[charisoudis]

Adding to @LokeshJatangi 's solution, one has to modify the config imports as well.

[bibekyess]

#11913 provides a simple fix to this.