search

UttaranB127 / STEP

Spatial Temporal Graph Convolutional Networks for Emotion Perception from Gaits
https://gamma.umd.edu/step
MIT License
69 stars 12 forks source link

Some questions about the dataset #11

Closed cvychen closed 9 months ago

cvychen commented 3 years ago

Hello sir, I have found that the scale of dataset depicted in the paper is different from the one in the website. In the paper, there are 5227 = 4227(real) + 1000(synthetic). However, I get 6177=2177(real) + 4000(stnthetic) in the website. Please tell me the reason.

UttaranB127 commented 3 years ago

Hello, thanks for your question. We had to make changes to the dataset since its original upload (primarily due to logistical issues and expiring storage permissions) and as a result, there are fewer of the originals currently available. I hope this answers your question.

cvychen commented 3 years ago

Thanks so much for your reply! I get it.

cvychen commented 3 years ago

Hello Uttaran, I want to know whether should I translate the labels [angry,happy,neutral,sad] to be [0,1,2,3] when I use the synthetic dataset. Because when I open the real dataset label file, I have found the rule is not suitable. As following words you see: /001_Angry_positions.csv:1 /001_Happy_positions.csv:3 /001_Neutral_positions.csv:1 /001_Sad_positions.csv:3; /002_Angry_positions.csv:1 /002_Happy_positions.csv:3 /002_Neutral_positions.csv:3 /002_Sad_positions.csv:3; /003_Angry_positions.csv:1 /003_Happy_positions.csv:0 /003_Neutral_positions.csv:3 /003_Sad_positions.csv:2; I hope to hear your explanation。 Thanks！

UttaranB127 commented 3 years ago

Thanks for your observation! The labels map one-to-one to the following: [0, 1,2 3] -> ['Angry', 'Neutral', 'Happy', 'Sad']. You can find the mapping in multiple places in our code, e.g., in classifier_stgcn_real_only/main.py, line 75.

The file names actually indicate the intended emotions of the subjects whose gaits constitute our dataset. The labels are the perceived emotions that labelers assigned after looking at the gait data. We use the perceived emotions in our experiments.

cvychen commented 3 years ago

Thanks for your explanation! Have you run the model on the modified dataset as the dataset is different from the one depicted in your paper? Could you please tell me the accuracy you get and the parameter you set during the training time if you have done it?

UttaranB127 commented 3 years ago

No sorry, we have not tested this method on other datasets yet.

cvychen commented 3 years ago

Thanks for your kind reply!