4 ms·
End-to-End Referring Video Object Segmentation with Multimodal Transformers
- andrewflnr 5y agoI'm having trouble understanding the problem being solved here. It looks like we're trying to figure out which pixels in each frame of video correspond to an object named in some accompanying text. I'm not sure if the text is an input or output, though, or what other parameters might be in there (timestamps/intervals of the video?).
- EvgeniyZh 5y agoThe input is video + text query. The output is mask of the object referred by the textual query in each input frame (possibly empty in some frames).
- andrewflnr 5y agoAwesome.
- wdb 5y agoHandy as base for rotoscoping :)