Self-supervision fromepipolar projection can improve robustness and accuracy of the 3Dperception and localization of vision-based robots . The rigidprojection computed by ego-motion cannot represent all scene points, such aspoints on moving objects, leading to false guidance in these regions . We propose an Attentional Separation-and-AggregationNetwork (ASANet) which can learn to distinguish and extract the scene’s static and dynamic characteristics via the attention mechanism . We further propose anovel MotionNet with an ASANet as the encoder, followed by two separatedecoders, to estimate the camera’s . scene’s dynamic motionfield .

Author(s) : Feng Gao, Jincheng Yu, Hao Shen, Yu Wang, Huazhong Yang

Links : PDF - Abstract

Code :

https://github.com/oktantod/RoboND-DeepLearning-Project




Keywords : dynamic - scene - propose - attentional - asanet -

Leave a Reply

Your email address will not be published. Required fields are marked *