limaosen0 / AS-GCN

The model architecture of AS-GCN (for human action recognition)
BSD 2-Clause "Simplified" License
290 stars 71 forks source link

Really low accuracy on NTU RGB D 120 #13

Closed Anirudh257 closed 5 years ago

Anirudh257 commented 5 years ago

Firstly, thanks a lot for making the code public. I am trying to train AS-GCN on the cross-setup setting as described in NTU RGB D 120. For performing this experiment, I am doing the following steps: 1) Making changes in the preprocessing file to create new data with 120 classes and even setups for training and odd setups for testing.

2) I make the changes in the yaml config file and following the same procedure as given for the NTU RGB 60.

I am getting decent accuracy for the Cross-Subject protocol but really low on Cross setup. What can be the reason for this?

czhaneva commented 5 years ago

May I ask how long did it take to train the model ?

Anirudh257 commented 5 years ago

@Roy790422795 It takes almost 2 days to train the model on a single GPU on my university cluster. The batch-size is too large.

dangao250 commented 4 years ago

Can I have the dataset, please

Anirudh257 commented 4 years ago

@dangao250 You can go to the NTU 120 website and request for the link to download the dataset.

dangao250 commented 4 years ago

@dangao250 You can go to the NTU 120 website and request for the link to download the dataset.

I applied on the official website but haven't replied for a month

Anirudh257 commented 4 years ago

@dangao250 I also faced the same issue. You should mail the authors listed on the paper. They mailed promptly and I was able to download the dataset.

FesianXu commented 3 years ago

@Anirudh257 Hi, Have you tried it on NTU 60? How about the performance of the model on the NTU RGBD 60 ? I want a comparision on my own database. Thx

Anirudh257 commented 3 years ago

@FesianXu I haven't trained on NTU-60 but I think that it should get done faster than NTU-120 in a day. While comparing on your own dataset, ensure that the graph is properly constructed and the ordering is correct.

FesianXu commented 3 years ago

@Anirudh257 Thanks for your response. I will try.