Why split UCF-101 train/test ?

kenshohara / 3D-ResNets-PyTorch

3D ResNets for Action Recognition (CVPR 2018)

MIT License

3.89k stars 931 forks source link

Open michaelHaha opened 4 years ago

michaelHaha commented 4 years ago

It does not make sense to split UCF-101 origin train/test partitions into your train/val/test split, which makes your results incomparable to others.

michaelHaha commented 4 years ago

I will fix it by myself.

HuangZuShu commented 3 years ago

@michaelHaha I think so as you. I don't know why demo use the same data in val and test ? Have you solve this problem?

sebyo commented 6 months ago

@michaelHaha did you find a solution ?