ningma-ai / dastm Goto Github PK

Hello, thanks for your excellent work and the elegant code released! Is the accuracy reported in the paper the validation accuracy? I re-run the code and found that the validation accuracy is almost identical to the paper, but the test set accuracy is somewhat lower.

请问文章的附录去哪里可以找到？

Hi, your work is very innovative! May I ask if you have considered the two key points for action recognition in the process of processing the data set: different individuals perform the same/different actions, and the same individuals perform the same/different actions under different perspectives? I look forward to receiving your reply, thank you very much.

Hi, thanks for your excellent research work. Is there any possibility that you could upload the msg3d and 2s_agcn network python files which could be used in this project?
Thank you so much.

Your work is excellent! If I want to train "5-way-5-shot", what should I do? I refer to the parameter setting of "5-way-1-shot" and set parameter backbones use STGCN, data use NTU-T, SA=1, reg=0.1, num_support_tr=5, num_support_val=5. The accuracy（test acc: 0.8649239775717259） is very different from the accuracy reported in the paper. May I ask if you have encountered similar situations in the training process. Looking forward to your reply.

Hello, Thank you for your code. I download the data. But after loading the ntu-T dataset, the info is printed as follows:

_sample_num in train 2320
class 80
data_path :/data/Disk_B/action_data/self_ske/ntu120/NTU-T
sample_num in val 580
n_class 20
data_path :/data/Disk_B/action_data/self_ske/ntu120/NTU-T
sample_num in test 580
n_class 20

It seems 29 samples for each class. Can you double-check this info as it should be accurate in the paper?

您好作者，看见您实验了了ST-GCN 2s-AGCN MS-G3D三个backbone，请问如果我想实验自己的backbone，该在项目的何部分进行改动呢。我的backbone类似于CTR-GCN，其由2s-AGCN改版而来

Hello, many thanks for the beautiful code you've shared and your outstanding effort. I plan to train the model on other datasets, but I don't know how the three files train_data.npy, train_label.npy, train_frame.npy are obtained, thank you very much!