Beside the compilation of matcaffe which took me a lot of time, I had to remove the absolute paths in custom_toolboxes/scene_parsing_model/scene_parse.m. The interactive selection of the images was not working, thus I removed the call to select_image in replace_sky.m, and selected the image programatically (automatically select the first image). The code needs a mask and a parsing image for each RGB image took as input. No code is provided to generate those, which limits its use in other contexts.
If you want to contribute with another review, please follow these instructions.
Please consider to cut/paste/edit the raw JSON data attached to this paper.