search

microsoft / View-Adaptive-Neural-Networks-for-Skeleton-based-Human-Action-Recognition

View Adaptive Neural Networks for High Performance Skeleton-based Human Action Recognition
MIT License
127 stars 58 forks source link

contradiction between paper and code for translation #1

Closed sksenthilkumar closed 4 years ago

sksenthilkumar commented 4 years ago

In the published paper "View Adaptive Neural Network for High-Performance S-b HAR", the translation was described as subtracting the obtained translation values from the given point. But However in the implemented code (transform_rnn.py and transform_cnn.py), it looks like the translation values are added to given points. Is this a bug? or I Understood it wrong?

Thanks in advance.

shuidongliu commented 4 years ago

The translation values are learned by the networks and the values can be positive or negative. Therefore, it does not matter about using add or subtraction.