search

hkchengrex / Mask-Propagation

[CVPR 2021] MiVOS - Mask Propagation module. Reproduced STM (and better) with training code :star2:. Semi-supervised video object segmentation evaluation.
https://hkchengrex.github.io/MiVOS/
MIT License
127 stars 22 forks source link

Out of memory was reported when batch_size=1 when testing the model #31

Closed longmalongma closed 3 years ago

longmalongma commented 3 years ago

What parameters should be changed to reduce the memory footprint when the batch_size=1 is reported to be out of memory?

hkchengrex commented 3 years ago

What videos are you testing on?

longmalongma commented 3 years ago

What videos are you testing on?

I tested this on the youtube_vos dataset by default

hkchengrex commented 3 years ago

It should not run out of memory with a 11GB GPU.

longmalongma commented 3 years ago

It should not run out of memory with a 11GB GPU.

@hkchengrex I'm testing my modified model, which may have more parameters and more complexity

hkchengrex commented 3 years ago

Well then you would need to make your own changes. General advice can be found here: https://github.com/hkchengrex/Mask-Propagation/issues/8

longmalongma commented 3 years ago

Well then you would need to make your own changes. General advice can be found here: #8

@hkchengrex The easiest way is to increase the number mem_freq? Change 5 to 10?

hkchengrex commented 3 years ago

I have already given that suggestion in the linked issue above. So yes.

longmalongma commented 3 years ago

I have already given that suggestion in the linked issue above. So yes.

Ok, thank you very much!