AI
ROWBench: Do Video Models Render What the Program Specifies?
Publication: Hugging Face Daily Papers
Author: Not available in collected metadata
Source published: 30/09/2026, 20:00:00 UTC
Briefing created: 02/10/2026, 23:06:30 UTC
Based on a feed excerpt; full article content has not been acquired. Source evidence retrieved 02/10/2026, 03:18:25 UTC.
Briefings
PROWBench introduces 170 programmatically constructed episodes and 600 proxy videos to evaluate video models against explicit program specifications. Unlike existing benchmarks that focus on visual quality, PROWBench logs entity states and timestamped events to verify if generated videos match the observable consequences of program execution. The benchmark covers first- and third-person perspectives with synchronized multi-view observations for a subset of episodes, using two VLM-based metrics to measure Logic-Render Alignment and Interaction Success Rate.
Key points
- PROWBench introduces 170 programmatically constructed episodes and 600 proxy videos to evaluate video models against explicit program specifications.
- The benchmark evaluates adherence to the prescribed timeline and visual realization of timestamped engine-recorded events using two VLM-based metrics.
Context and limitations
Coverage is limited to 170 programmatically constructed episodes and 600 proxy videos, with synchronized multi-view observations available only for a subset of episodes.
Generated by Bittrees-hosted qwen35-2b sha256:aaf42c8b7c3c on 02/10/2026, 23:06:30 UTC. Based on publisher-provided feed evidence; may not cover the full original article. Check the source for details.
View archived briefing on IPFS
bafybeihqgyguxt6bwzwrdeqp3jq22rdh4vbdzsduqhrn2mpr3ltxeucqse