CVPR Daily - Thursday

UnweaveNet considers the novel problem of addressing the complexity of daily life captured by long untrimmed videos to unweave a video into its constituent activity threads . When someone wakes up in the morning and walks into their kitchen, they may first put the kettle on to make a coffee, then while it is boiling, put some bread in the toaster, then while that is toasting, start some washing up. The process begins with one goal but introduces new goals along the way, and the person’s focus switches back and forth as these tasks are interleaved, paused, and resumed until the final goal is reached . “ This is very different to how activity understanding models and methods have worked to date because the type of footage they’ve been dealing with hasn’t been so complex, ” Dima tells us. “ Increasingly, there’s a trend for capturing unscripted footage in data sets like EPIC-KITCHENS, which poses questions about how people understand videos and how we can parse or process this input. ” Dima Damen is a Professor of Computer Vision at the University of Bristol. She speaks to us about UnweaveNet, a collaboration between first author and Bristol PhD alumnus Will Price; Carl Vondrick from the University of Colombia; and Dima herself. UnweaveNet will be presented as a poster this afternoon. UnweaveNet: Unweaving Activity Stories 8 DAILY CVPR Thursday Poster Presentation