Unverified50% confidenceFactExact time
ConViS-Bench is a benchmark dedicated to evaluating concept-level video similarity, covering multiple concept dimensions such as activity, location, and objects, and requiring models to judge whether two videos agree on a specified concept.
1
Sources
50%
Confidence
Long-term
Relevance
10/20/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedVICE (Video Caption Entailment) is a method that turns cross-video comparison into a video-language entailment task through a dual-branch design, for concept-level video similarity estimation.66% similarUnverifiedVSI-Bench is a benchmark for video spatial intelligence, covering tasks such as object counting, relative orientation, path planning, and perspective transformation.65% similarUnverifiedThe scene graph serves as a generative constraint that guides the sampling of diverse reset states, thereby covering object poses and grasp modes that do not appear in the video.65% similarPartially VerifiedAgentic video understanding allows models to autonomously decide where to look, how long to look, which segments to skip, and to perform multi-step reasoning in conjunction with context.64% similarUnverifiedOVO-Bench and StreamingBench are evaluation benchmarks designed specifically for streaming/online video understanding, whose core feature is simulating 'watching while answering', where at time t the model can only see frames that have arrived before that point.64% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1074254API
curl https://kongchang.com/api/v1/knowledge/claims/1074254MCP
get_claim(id=1074254)