search

shlizee / Predict-Cluster

Repository for PREDICT & CLUSTER: Unsupervised Skeleton Based Action Recognition
Other
107 stars 23 forks source link

Train on 2D skeleton dataset #20

Closed ChenConnie closed 3 years ago

ChenConnie commented 3 years ago

Hello, Thank you for your excellent work. The code release the preprocess data on UCLA, UWA3D and NTU RGBD. It seems they all 3D skeleton data. Can this code work on 2D skeleton dataset like COCO? If it is, could you tell me how? Thank you again.

sukun1045 commented 3 years ago

We haven't tried it on the COCO dataset but we did successfully apply it to 2D animal trajectories. You can simply replace the input and try it yourself.

ChenConnie commented 3 years ago

OK. Thank you for your reply.

tctco commented 1 year ago

@sukun1045 Hi! I'm wondering how to apply your method to animal trajectories. Could you give me some guidance on how to do this? Thanks!

sukun1045 commented 1 year ago

Hi, I believe you can perform similar training for animal trajectories. For example, you can prepare your animal trajectories data as a sequence with shape (time, Feature_dim) and train an encoder-decoder model with a weak decoder and simple reconstruction loss as described in the paper. Once you are done with the training, you can investigate the encoder's final state to see whether it contains any meaningful patterns.