AWhat changed
Mapped varying image resolutions to variable visual-token counts and decomposed rotary position encoding across text, image and video axes within a unified visual-language model.
Scientific scope
Mapped varying image resolutions to variable visual-token counts and decomposed rotary position encoding across text, image and video axes within a unified visual-language model. Original QwenVL is inherited; dynamic tokenization/M-RoPE is the scoped change. Native resolution is constrained by practical resizing/token budgets, not unlimited visual resolution or proven human-like perception. Current repository redirects to Qwen3VL, whose features are not backdated.
Evidence
The linked original sources describe the stated contribution and its evidence limits. WRS has not independently rerun the experiments or certified a proof. Original QwenVL is inherited; dynamic tokenization/M-RoPE is the scoped change. Native resolution is constrained by practical resizing/token budgets, not unlimited visual resolution or proven human-like perception. Current repository redirects to Qwen3VL, whose features are not backdated.
Contributors
- Source-listed authors and technical contributors · SCIENCE: Source-listed authors or explicitly credited technical contributors for the cited versions. Family entries retain source-specific attribution; a combined list is not one original-paper byline or equal individual work. Names alone do not establish resolved identities.
- Qwen2-VL: Enhancing Vision-Language Model's Perception of the World at Any Resolution
- Wang et al.: dynamic-resolution vision tokens and multimodal RoPE — author/institutional project
Provisional WRS assessment, edition 2026-10-16.1. Independent expert calibration is pending. This grade does not certify correctness or assign ownership.
Full scientific record · Browse contribution grades · Request a correction