LemonQC

Results 18 comments of LemonQC

Well, I found this is a two-stage process. I utilize ffmpeg to set all the videos at a specific speed, such as 15fps. Then, I use HCRN to sample frames...

@doc-doc I first transfer the video to a certain fps and then extract. Does this work?

ok, many thanks. Is the model in nextqa for question features the final model? Or I still need to train? I just utilize the model to extract question feature while...

你是几个卡训练的,也可能跟机器有关

Have you solved?

I change the num_apperance={aff:1 ..}, The results are: mIoU: 0.23582979498476714, Acc: 0.4684787326388889, mAcc: 0.4987481809129902 mESM: 0.1768201153706607, mCR: 0.2348148148148148