search

hkchengrex / XMem

[ECCV 2022] XMem: Long-Term Video Object Segmentation with an Atkinson-Shiffrin Memory Model
https://hkchengrex.com/XMem/
MIT License
1.76k stars 192 forks source link

Processing videos in batches #129

Closed WangYixuan12 closed 12 months ago

WangYixuan12 commented 12 months ago

Hi Rex, thank you for the amazing work! I am using XMem for multi-view VOS. Currently, I create multiple processors and process videos in series. I am wondering whether it is possible to take in multiple videos as batches and process them in parallel to speed up. Thank you for your help!

WangYixuan12 commented 12 months ago

I guess a more concrete question is whether step function can take in images in batch

hkchengrex commented 12 months ago

Thank you for your interest.

This is not implemented. As a side note, Cutie would probably run faster and better.

WangYixuan12 commented 12 months ago

Thank you for your reply! Do you have any quantitative comparisons between Cutie and XMem? I notice that XMem has ~20FPS. What is the expected FPS for Cutie?

hkchengrex commented 12 months ago

Paper: https://arxiv.org/abs/2310.12982 Around 35. FPS depends on the resolution/the number of objects just like XMem.

WangYixuan12 commented 12 months ago

Got it! Thank you for your explanation!