yolov10
yolov10 copied to clipboard
THU-MIG
📝 Add Hugging Face Demo and Model Pages to README
HuggingFace Demo(Zero A100): https://huggingface.co/spaces/kadirnar/Yolov10
HuggingFace Model Page: https://huggingface.co/kadirnar/Yolov10
Colab Demo: https://colab.research.google.com/github/roboflow-ai/notebooks/blob/main/notebooks/train-yolov10-object-detection-on-custom-dataset.ipynb Thanks @SkalskiP
@kadirnar Thanks for your great efforts! We found that the prediction results are not the same in our local environment and the HuggingFace demo. After investigation, we found that this is caused by the incorrect input format. As shown in the document for prediction:
https://github.com/THU-MIG/yolov10/blob/main/docs/en/modes/predict.md#inference-sources
For the numpy.array input, the model expects the input with the format of BGR. However, the gr.Image(type='numpy') component will output the numpy.array with the format of RGB, which results in the incorrect predictions. Replacing gr.Image(type="numpy", label="Image") with gr.Image(type="pil", label="Image") can address this issue by passing the correct PIL format with RGB. Could you please fix this issue in the HuggingFace demo? We'd appreciate it very much. Thank you!
@kadirnar Thanks for your efforts! But you seem to have changed it in the wrong place. Could you please replace gr.Image(type="numpy", label="Image") in this line https://huggingface.co/spaces/kadirnar/Yolov10/blob/main/app.py#L52 with
gr.Image(type="pil", label="Image")? Thanks a lot!
I fixed it. Thank you for the detailed explanation ❤️