search

fabiocarrara / mocap

Code to reproduce experiments in 'LSTM-based real-time action detection and prediction in human motion streams'
29 stars 7 forks source link

How can i find HDM05 dataset mapping relationship? #4

Open yjb767868009 opened 2 years ago

yjb767868009 commented 2 years ago

I really admire your work for mocap annotation. I get the original dataset of HDM05 and the dataset with annotation you provided, but I can't match them. Please help me, thanks.

fabiocarrara commented 2 years ago

Hi @yjb767868009 , this might not be trivial (that's the reason we provide already segmented data). In the Dataset subsection of our paper, you can find some details and references to previous work defining those splits. If you can't figure it out, I suggest contacting Dr. Jan Sedmidubsky; his lab carried out data preparation and can help you better than me.

yjb767868009 commented 2 years ago

I appreciate the segmented data you provided. But I want to annotate on HDM05 original dataset, so I need the relationship between the two datasets. Thank you for your help.

xsedmid commented 2 years ago

Attached is the text file containing the mapping between our data and the original HDM05 data. The file has two columns separated by a semicolon: the first column corresponds to our synthetically-assigned sequenceID and the second column to the corresponding original HDM05 sequenceID.

In particular, the HDM05 dataset contains long skeleton sequences in which short actions are annotated (labeled). We work with both short actions (used for the action classification task) as well as long sequences (used for the action detection task).