hkchengrex / XMem

[ECCV 2022] XMem: Long-Term Video Object Segmentation with an Atkinson-Shiffrin Memory Model

Home Page:https://hkchengrex.com/XMem/

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Validation Loss while Training

abdksyed opened this issue · comments

Is there a way to get validation loss while training.

I want to fine-tune the model on my dataset, so I am only training for stage-3 for 3000 iterations with batch_size of 8 and 16 num_frames.

But I want to see validation loss or validation IoU on the test set, while training. My concern is maybe the training may overfit, for now I am saving weights after every 50 iterations and trying to see IoU for each weight by running inference.

I think since the model will have memory, it is difficult to do inference and validation metrics during training, but I wanted to know is there any way to do so?

It can definitely be implemented. As you said, it would involve memory updates so the implementation might be a bit hairy.

Is there a way to get validation loss while training.

I want to fine-tune the model on my dataset, so I am only training for stage-3 for 3000 iterations with batch_size of 8 and 16 num_frames.

But I want to see validation loss or validation IoU on the test set, while training. My concern is maybe the training may overfit, for now I am saving weights after every 50 iterations and trying to see IoU for each weight by running inference.

I think since the model will have memory, it is difficult to do inference and validation metrics during training, but I wanted to know is there any way to do so?
Hello, may I ask if you have trained 3000 rounds, num_ How is the effect of setting frames to 16? Have you made any adjustments to the p-value in Celoss? May I ask about the approximate quantity of your dataset? Thank you.