EvidenceConnect boosts multimodal perception without zooming, sweeping 8 benchmarks
inclusionAI · hf · 2026-10-28
A team from Shanghai Jiao Tong University, Ant Group and Fudan releases Zooming without Zooming v2, introducing EvidenceConnect: synthesized training data that strengthens visual evidence connection, improving spatial perception and multimodal reasoning without actual zooming.
- Benchmarks in the paper show the 4B/9B/27B variants beating GPT-4o, Gemini and other baselines on ZoomBench, PerceptionBench, MMSI-Bench, SPAR-Bench, MileBench, ERQA, ViewSpatial and MMStar.
More from Multimodal
- FastH3 workflow runs locally in ComfyUI to generate a cocktail video clip — bennash · 2026-10-28
- Playcanvas proposes SOG v3: 4-band SH, opacity above 1, per-gaussian blind data — willeastcott · 2026-10-28
- Voxtral Mini Transcribe 3 ranks #2 on Voice Arena diarization leaderboard at 40.4% DER — MistralAI · 2026-10-28
- Imaginary Tokens in Midjourney v8.2: Four Made-up Words, Four Surreal Styles — LudovicCreator · 2026-10-28
- Camera-follows-action prompt trick for thrill-ride AI video generation — umesh_ai · 2026-10-28
- Cinematic scene transitions made with Google Gemini and Flow, prompts shared — michaelrabone · 2026-10-28