Is the pre trained model you provided already trained data? I have a significant loss in the validation set testing after loading.

[chongchongjjj]

Is the pre trained model you provided already trained data? I have a significant loss in the validation set testing after loading.

[wenbowen123]

The validation loss cannot fully reflect the grasping performance. It's suggested to run the grasping to test.

[chongchongjjj]

Sorry, I didn't make it clear. What I wanted to express was the training of nunocs. After loading the pre trained model you provided, the initial cross entropy loss is approximately over 20.

[wenbowen123]

Can you show the code you used to compute the loss and did you switch to eval mode for the model?
And have you tried running the grasping demo? If that works, it means the weights are loaded correctly.

[chongchongjjj]

I just modified the train_ In the model loading section of the nunocs.py file, I loaded your pre trained nunucs model and annotated the code for the training loop section. I only used val loop to help me check the loss situation. What I want to reproduce is your other paper (behavioral cloning). Here, I only want to call the nunucs network and not run the code for the capture section.