search

yjxiong / tsn-pytorch

Temporal Segment Networks (TSN) in PyTorch
BSD 2-Clause "Simplified" License
1.07k stars 308 forks source link

training tsn on kinetic400 is ok, but validating is very low. do you test your model on the kinetics400 data set #114

Closed LiuChaoXD closed 3 years ago

LiuChaoXD commented 4 years ago

I am training tsn model on kinetics400 data. when the epoch is [30/100], I get the accuracy 30% on training set. However, the accuracy of validate set is 5%.

yojayc commented 4 years ago

Did you solve the problem?

My training accuracy is also quite low

LiuChaoXD commented 4 years ago

Did you solve the problem?

My training accuracy is also quite low

I think the problem is caused by the kinetics400 data set (probably). I test the tsn model with their released trained weights on my kinetics400, unfortunately the accuracy is very low too. I have no the original Kinetics400 data set, such that I don't know the problem is caused by the data sets or the model.

cy-kui commented 3 years ago

hello, i need the kinetics dataset, can you send me that? thank you! 2115996745@qq.com