I want to use this project on "Something-something"

fabienbaradel / object_level_visual_reasoning

Pytorch Implementation of "Object level Visual Reasoning in Videos", F. Baradel, N. Neverova, C. Wolf, J. Mille, G. Mori , ECCV 2018

172 stars 20 forks source link

I want to use this project on "Something-something" #6

Open CxuS opened 5 years ago

CxuS commented 5 years ago

here have something questions. 1、I want to use this project on "Something-something", but i haven't this mask data. 2、The dataset of "something-something" with 174 classes in 20bn. but I find only 157 classes in your paper. so I don't know that whether "SS" in 20bn .

fabienbaradel commented 5 years ago

Hi @CxuS, I encourage you to work on Something-Something V2 because the video resolution is higher (minimum high/weight = 256). I have been using Something-Something v1 for this project and the object detection predictions made by MaskRCNN were not perfect compared to VLOG for example. You can use the torchvision pipeline for running maskrcnn on the videos. Thanks poiting out this mistake, you are right there are 174 I did a mistake in the paper. I will update this section.

CxuS commented 5 years ago

Thank you for your reply and your advice. Best regards !