Section: New Software and Platforms

M-CNN: Weakly-Supervised Semantic Segmentation using Motion Cues

Participants : Pavel Tokmakov, Cordelia Schmid, Karteek Alahari.

This is a public implementation of the method described in [23]. It includes a framework for integrating motion cues into training a deep network for weakly-supervised semantic segmentation, code for data preprocessing and trained models that correspond to the results reported in the paper. Our code is built on top of DeepLab https://bitbucket.org/aquariusjay/deeplab-public-ver2 extension of the Caffe deep learning framework http://caffe.berkeleyvision.org.