Inference with pre-trained model?

[fipeop]

I am trying to run inference in an external dataset using the pre-trained weights. However, both test.py and pred_pseudo.py refer to resume checkpoint: ./out/Nvidia/old_fold0/aepoch_500.pth

Is this on purpose or are you planning to upload the checkpoint?

[kretes]

Have you downloaded checkpoints linked in https://github.com/ljwztc/CLIP-Driven-Universal-Model#model and put correct one at the path it's expecting it?

That's not an official answer, just what I think is required.

[fipeop]

Yes, those are the weight files .pt, but the --resume option in both test.py and pred_pseudo.py seems to refer to a .pth checkpoint file. As the inference script is written now, you need it seems.

[ljwztc]

Hi, @kretes is right. Download the checkpoint from link and resume the pth using --resume. ./out/Nvidia/old_fold0/aepoch_500.pth is just used in our server.

[fipeop]

Yes, it worked with the new checkpoint and some modification of the loading approach.