search

dulucas / siMLPe

A Simple Baseline for Human Motion Prediction
MIT License
134 stars 20 forks source link

Auto-regressive motion prediction for long periods #5

Closed hygenie1228 closed 1 year ago

hygenie1228 commented 1 year ago

Thank you for sharing the interesting work.

I want to inference the network for long-term (target_length > 25) WACV paper denotes "During testing, we apply our model in an auto-regressive manner to generate motion for longer periods" However, I couldn't find the auto-regressive code in this repository. Is there any way to generate long-term motions (target_length > 25)?

dulucas commented 1 year ago

Hi, the auto-regressive test_function is already in the code, for example here. Our model is trained to output 10 future frames, and the target length in our code is 25, thus we applied auto-regressive test to get the target length. You can get longer prediction by changing the config.motion.h36m_target_length_train in the config.py. Hope it helps