Joon

It seems that the importance sampling code part is wrong.

2

https://github.com/ikostrikov/pytorch-trpo/blob/e200eb8a23b3c7941a0091efb9750dafa4b23cbb/main.py#L108-L119 The fixed log prob part of the line and the "get_loss" function part are exactly the same. The two parts are executed consecutively so that the two values ("fixed_log_prob",...

VE scheduler for flow matching.

1

Thank you for sharing your great implementation! I would like to ask whether the VE setting works well. Now, I am trying to implement the flow matching of the VE...

The dimension of the input data

1

Thank you for sharing your great work. In 3D case, for employing DeepONet, should I flatten the input function? Something like : (B, C, H, W, D) -> (B, L)?...

Add the data serialization factor in COCO-JSON-LOADER

2

This is for the issue #307 . (Closes #307 ) When I finetuned the sam3 with a large dataset (like larger than 1M 2D images), I faced the memory leak...

Memory Leakage Problem?

2

When I tried to fine-tune the model with a large annotation set (about 2M), memory usage gradually increased during the training process. Ultimately, the process was exploded.