ResearchBenchmarksMulti-conditioned Retrieval on Stanford40 Action + LocationFollow58.5mAPCLAY46.74849.79952.8555.901Apr 13, 2026Evaluation ResultsMethodMethodLinksmAPCLAYBackbone=CLIP-LBackbone=CLIP-L2026.0458.5CLAYBackbone=CLIP-BBackbone=CLIP-B2026.0457.9GeneCIS†2026.0455.5GeneCIS2026.0447.2