HarborYuan/ovsam

FileNotFoundError

Closed this issue · 6 comments

CVHub520 commented

Hi, could you please provide the download link for RN50x16_CocoOVDataset.pth? I couldn't find the relevant download link.

HarborYuan commented

You may generate it by yourself:

bash tools/dist.sh gen_cls seg/configs/ovsam/ovsam_coco_rn50x16_point.py 8

Please also refer to the details here:

https://github.com/HarborYuan/ovsam#-training

CVHub520 commented

Resolved, thanks for the response.

BTW, could u please provide an inference demo if possible? Currently, it seems that only training and testing code is provided.

HarborYuan commented

Resolved, thanks for the response.

BTW, could u please provide an inference demo if possible? Currently, it seems that only training and testing code is provided.

Please refer to the second item of the Usage in README.md

CVHub520 commented

Thank you for your feedback. Currently, I can use gradio for inference, as you mentioned.

However, while I attempted to convert the model using torch.onnx.export, the resulting ONNX file was always empty. Additionally, a series of seemingly random numbers were continuously printed.

(1,384,.,.) = 
 0.001 *
  7.4430
...
(255,384,.,.) = 
 0.01 *
  1.6353

(256,384,.,.) = 
 0.01 *
  2.2419
[ torch.FloatTensor{256,384,1,1} ]
HarborYuan commented

Hi @CVHub520 ,

I don’t have much experience in model deployment. Could you please have a try on MMDepoly? I guess they provide some tools to deploy models in the OpenMMLab ecosystem.

Please let me know if you have any other question.

haoyi-duan commented

Hi, @CVHub520, any update on the inference model deployment?