pretrained model

[ak9250]

can you share your pretrained model and testing code

[vincent-thevenin]

Pasted a link in the ReadME to the weight files on my drive. Testing code coming soon.

[ak9250]

@vincent-thevenin thanks looking forward to the test code

[EricHallahan]

It looks like the file will not unarchive.

$ tar -xvf model_weights.tar
  tar: This does not look like a tar archive
  tar: Skipping to next header

[vincent-thevenin]

You don't need to unarchive it. You can use the path to the file as is.

[ak9250]

@vincent-thevenin currently inference is through webcam, is there a test for single image or multi image inputs?

[iyah4888]

It seems that webcam_inference.py requires "e_hat.tar", but it is not documented in anywhere.
Could you give some context and share the pre-trained model of e_hat?
Thanks!

[vincent-thevenin]

@ak9250 The class methods for finetuning are already coded in. I still need to write the code for training in finetuning mode, I will work on this.

[vincent-thevenin]

@iyah4888 "e_hat.tar" is an output of embedder network. You can run inference on the embedder on any picture and get an e_hat.

I will add some lines to explicitely create it thanks for reminding me!

[vincent-thevenin]

@ak9250 You can now train the network in finetuning mode on a single video or image. You have to run finetuning_training.py and change the paths in the file to the video or images you want to use.

[vincent-thevenin]

@iyah4888 You can now run embedder_inference.py to create your own embedding.