Spaces:

nikigoli

/

countgd

like 26

Running on T4

Hi, thank you for your message and interest! Please refer to this GitHub issue, which includes how to run the app in docker. You can also pip install spaces as described here if you still run into this issue. Please let me know if you have any further questions (:

Hi, if you are having the issue pasted below, this is a new issue that appeared a couple of days ago. I am currently in the process of investigating and fixing it:
ERROR: Exception in ASGI application
Traceback (most recent call last):
File "/home/user/.pyenv/versions/3.10.15/lib/python3.10/site-packages/pydantic/type_adapter.py", line 270, in _init_core_attrs
self._core_schema = _getattr_no_parents(self._type, 'pydantic_core_schema')
File "/home/user/.pyenv/versions/3.10.15/lib/python3.10/site-packages/pydantic/type_adapter.py", line 112, in _getattr_no_parents
raise AttributeError(attribute)
AttributeError: pydantic_core_schema

During handling of the above exception, another exception occurred:
etc...

Hi, I fixed this issue in the commit: 852f01bd33bf2213cf2ab71e7355317f56e404cb by updating the requirements.txt with fastapi==0.112.2. It seems this is a common issue with Gradio. I found the solution here: https://github.com/gradio-app/gradio/issues/9275.

hi @nikigoli , thanks a lot for sharing both new research, code, and demo scripts. Very kind of you guys :)
I've a question, could I know, can countgd be exported to Torchscript? or have you tried to convert it to other format (ONNX/Torchscript/Openvino etc)?

The reason for me to ask is that I've got an error where I can't save the converted Torchscript model due to "MultiScaleDeformableAttnFunction".

Below is the log of torchscript conversion:

RuntimeError:
Could not export Python function call 'MultiScaleDeformableAttnFunction'. Remove calls to Python functions before export. Did you forget to add 

@script
	 or @script_method annotation? If this is a nn.ModuleList, add it to __constants__:

I have not had a chance to look into this yet. However, if there is a solution for Grounding DINO, I would expect there to be a solution applicable to CountGD. I see that this is a recent request for Grounding DINO: https://github.com/IDEA-Research/GroundingDINO/issues/363.
It seems someone has done this here: https://github.com/IDEA-Research/GroundingDINO/issues/178 . I will come back to this as soon as I get a chance, but if you find a solution in the meantime or draw any insights, feel free to mention them here.

Hi, can you share the demo implementation for single image inference as shown in this space. I find it difficult to figure out the command the countgd pipeline using grounding Dino while using it for inference on pretrained weights 1.2Gb from drive

Hi, the code is in count_main available here: https://huggingface.co/spaces/nikigoli/countgd/blob/2bb3f9251ec8101bf6cdefe83e37f4378b7c196f/app.py#L346 . The weights used in this space are the same as the weights from drive.