Creating Whole Garden of forking path dataset as single future

[junweiliang]

Creating Whole Garden of forking path dataset as single future

[JunweiLiang]

The "trained on simulation" experiment refers to the model trained on "anchor videos", which were reconstructed from the real-world ActEV videos. See here.
Please follow the training instructions from here to reproduce the main experiments if that's what you want.

[rameezrehman83]

Thank you for your reply. I want to reproduce the results with "anchor videos". I got a question regarding the preprocessing of "anchor videos". How do you use the "anno_scene" for "anchor videos" as I don't get them after running "get_prepared_data.py". In in "get_prepared_data.py" line 71,72, "anno_scene" is mentioned but is commented:
#we will have scene segmentation feature for each needed frame
#args.scene_map_path = os.path.join(args.outpath, "anno_scene")
Can you please comment on how to get "anno_scene" for anchor_videos. Thank you

[JunweiLiang]

I have not prepared the instructions for "anchor videos" experiment for the CVPR 2020 paper. Could you look at the instructions from here, which is in the ECCV'20 paper where we do training with multi-view anchor videos, and see whether that could help you first? Thanks.