Research map / Vision-language and multimodal / Multimodal LLMs
The All-Seeing Project V2: Towards General Relation Comprehension of the Open World
Weiyun Wang, yiming ren, Haowen Luo, Tiantong Li, Chenxiang Yan, Zhe Chen, Wenhai Wang, Qingyun Li, Lewei Lu, Xizhou Zhu, Yu Qiao, Jifeng Dai
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 changed | Where | Measured | Effect |
|---|---|---|---|
| add relation conversation (AS-V2) data to training (ablation) | ASMv2 · CRPE | overall accuracy: 55.3 -> 64.5 | Improved |
| add relation conversation data to training (ablation) | ASMv2 · MME | score: 1554.6 -> 1621.0 | Improved |
| add relation conversation data to training (ablation) | ASMv2 · MM-Vet | score: 42.2 -> 41.3 | Mixed |
| add relation conversation data to training (ablation) | ASMv2 · VCR | Q->AR accuracy: 77.0 -> 78.4 | Improved |
| add two-stage training (image-level stage 1 first) (ablation) | ASMv2 · MME | score: 1527.6 -> 1621.0 | Improved |
| add two-stage training (image-level stage 1 first) (ablation) | ASMv2 · MM-Vet | score: 35.6 -> 41.3 | Improved |
| add two-stage training (image-level stage 1 first) (ablation) | ASMv2 · CRPE | overall accuracy: 64.7 -> 64.5 | No clear effect |
| add two-stage training (ablation) | ASMv2 · RefCOCO/+/g average | accuracy: 86.1 -> 87.4 | Improved |
| include caption annotations in the GPT-4V prompt for data generation | AS-V2 data construction · data quality | hallucination in generated data | Improved |
| single-task fine-tune on VCR | ASMv2 · VCR validation | Q->AR accuracy: 78.4 -> 79.4 | Improved |
| train on relation conversation to predict specific predicates | ASMv2 · PSG Predicate Classification | mR@100 versus R@100: mR@100 44.5 vs VCTree 23.8; R@100 32.6 vs VCTree 53.1 | Mixed |
| open-ended generation of scene graphs with ASMv2 instead of TextPSG | Open-ended SGG · PSG | Recall: 5.5 -> 14.2 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- VILA-13B — VILA: On Pre-training for Visual Language Models (CVPR 2024) · better on 3 of 4 shared results
- Ferret-13B — Ferret: Refer and Ground Anything Anywhere at Any Granularity (ICLR 2024) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- ART: Adaptive Relation Tuning for Generalized Relation Prediction (ICCV 2025 · similarity 0.90)
- Vision-Language Interactive Relation Mining for Open-Vocabulary Scene Graph Generation (ICCV 2025 · similarity 0.89)
- Relation-aware Hierarchical Prompt for Open-vocabulary Scene Graph Generation (AAAI 2025 · similarity 0.87)
- Relation-R1: Progressively Cognitive Chain-of-Thought Guided Reinforcement Learning for Unified Relation Comprehension (AAAI 2026 · similarity 0.87)
- Synthetic Visual Genome (CVPR 2025 · similarity 0.86)
- ChainMPQ: Interleaved Text-Image Reasoning Chains for Mitigating Relation Hallucinations (ICLR 2026 · similarity 0.86)
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.
