Jie Lei 雷杰

@jwyang So what makes your accuracy lower than the reported one? I used maskrcnn-benchmark code to train/test the same splits, only got 2.24% mAP ( IoU 0.5).

Hi @avinashsai, I don't have an exact number, but should be around 12 hours.

The bottleneck is mostly on data loading (disk) and video decoding (cpu). You might want to see if you are bottlenecked by these.

Hi @Justin-TXX, possibly because the original video files we used to generate the lmdb file are broken. These files are only a small part of the files, you can simply...

Hi Ye, Yes, this looks like a typo.

Hi @prote376, We have the TGIF-QA frameQA config available here: https://github.com/jayleicn/ClipBERT/blob/main/src/configs/tgif_qa_frameqa_base_resnet50.json. If I remember correctly, we trained with this config on 4 GPUs. If you are using a different number...

I have added my coauthor Linjie @linjieli222 who conducted this experiment. Hi Linjie, Could you help us to verify the configurations? Thanks! Best, Jie

Hi @datar001, By default, we trained on 4-8 V100 GPUs. We have not tried training on a single 1080Ti. For this warning, it is expected due to the use of...

Hi @Zoe-Ziyi, There is no need to backup checkpoints, our code is only intended to backup the source code. The checkpoint files are rather large ~1-2GB. You should probably move...

Hi @angaiei007, please follow https://github.com/openai/CLIP#usage to extract the text features, this specific line below deals with text feature extraction: ```python text_features = model.encode_text(text) ```