AI Paper Atlas

Research map / Vision-language and multimodal / Multimodal LLMs

VersaViT: Enhancing MLLM Vision Backbones via Task-Guided Optimization

Yikun Liu, Yuan Liu, Shangzhe Di, Haicheng Wang, Zhongyin Zhao, Le Tian, Zhou Xiao, Jie Zhou, Jiangchao Yao, Yanfeng Wang, Weidi Xie

ECCV 2026 · Vision-language and multimodal › Multimodal LLMs · approach: Instruction-tuned multimodal assistants · 5 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
apply VersaViT multi-task post-trainingQwen2-VL-ViT · OpenCompass averageaverage score: 64.8 -> 66.4Improved
apply VersaViT multi-task post-trainingQwen2-VL-ViT · ADE20k linear probemIoU: 33.6 -> 49.6Improved
apply VersaViT multi-task post-trainingQwen2-VL-ViT · KITTI linear probeRMSE: 3.735 -> 3.136Improved
add VQA and captioning training only (ablation)Qwen2-VL-ViT · ADE20k linear probemIoU: 33.6 -> 32.0Hurt
add depth task to VQA and captioning (ablation)VersaViT · NYUv2 linear probeRMSE: 0.557 -> 0.495Improved
add segmentation task to VQA, captioning and depth (ablation)VersaViT · ADE20k linear probemIoU: 35.1 -> 49.6Improved
add segmentation task to VQA, captioning and depth (ablation)VersaViT · OpenCompass averageaverage score: 66.3 -> 66.4Improved
use balanced loss weights 0.33 each instead of 0.25/0.50/0.25 (ablation)VersaViT · OpenCompass averageaverage score: 65.7 -> 66.4Improved
apply multi-task post-training to Qwen2.5-VL-ViT (ablation)Qwen2.5-VL-ViT · ADE20k linear probemIoU: 37.2 -> 47.2Improved
apply multi-task post-training to Siglip2 (ablation)Siglip2-so400M-512 · OpenCompass averageaverage score: 59.3 -> 60.4Improved
use multi-task framework instead of more captioning/VQA data (35M) (ablation)Qwen2-VL-ViT · ADE20k linear probemIoU: 34.7 -> 49.6Improved
apply multi-task post-trainingQwen2-VL-ViT · SPair semantic correspondencerecall: 17.05 -> 26.99Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy