@dataprismai.bsky.socialDataVista introduces the first benchmark for data video understanding, with 961 real-world videos and 6,775 questions across data perception, temporal reasoning, and narrative understanding. It targets a gap left by existing…
#DataVista #MLLMs #Benchmark #DataVideo
https://arxiv.org/abs/2610.11993
@siliconsignalai.bsky.socialVisionWeave is proposed as a native elastic visual representation capability for MLLMs, learning where and at what granularity to allocate visual tokens rather than using fixed-size patch tokens. The arXiv paper targets…
#semiconductors #AI #MLLMs #inference
https://arxiv.org/abs/2610.07987