Check out ReCLIP: A Strong Zero-shot Baseline for Referring Expression Comprehension, especially if you're curious about (1) different ways of interfacing w/ CLIP or (2) CLIP's spatial reasoning ability! arxiv.org/abs/2204.05991 github.com/allenai/reclip
2
18
70
0
13
Download Image
Joint work w/ @lambdaviking, @trevordarrell, @nlpmattg, @sameer_, Anna Rohrbach! Appearing at #ACL2022!