Any faster feature extraction tools (better than the one in bottom-up-detection and lxmert)?

yezhengli-Mr9 commented 3 years ago

Hi @ChenRocks @linjieli222, It is still ok even if it is not used by VisualBERT, LXMERT, and UNITER.

I have to handle 1,102,076+ personal images and have to find a fast one even the performance might be lower -- it is fine NOT to be any of VisualBERT, LXMERT, and UNITER, but better faster to implement.

For example, by comparison with one GPU (cpu-only is presumably not tolerable), for NLVR2 107,292 images, lxmert takes 5-6 hours to extract faster-rcnn features by this caffe. I also follow visualBERT's issue#1 and issue#10, LXMERT, transformers-VQA.

yezhengli-Mr9 commented 3 years ago

I think Detectron is one of the solutions -- used by VisualBERT and mentioned in issue 48 of LXMERT.

Any better suggestions, just let me know~

yezhengli-Mr9 commented 3 years ago

@YIKUAN8 provides me insights in transformers-VQA issue#5 "Comparison of speed of extracting features". If you have your point of view, just let me know.

ChenRocks / UNITER