Research map / Vision-language and multimodal / Multimodal LLMs
VisualCOMET: Reasoning about the Dynamic Context of a Still Image
Jae Sung Park, Chandra Bhagavatula, Roozbeh Mottaghi, Ali Farhadi, Yejin Choi
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 person grounding (PG) tags (ablation) | Image + Event + Place · VisualCOMET val (text given) | CIDEr: 17.12 -> 18.27 | Improved |
| add person grounding (PG) tags (ablation) | Image · VisualCOMET val (no text given) | CIDEr: 5.63 -> 7.61 | Improved |
| add image input to event and place text (ablation) | Event + Place · VisualCOMET val (text given) | CIDEr: 15.73 -> 18.27 | Improved |
| add image input to event and place text (ablation) | Event + Place · VisualCOMET val (text given) | Acc@50: 33.06 -> 38.72 | Improved |
| add place input to image and event (ablation) | Image + Event · VisualCOMET val (text given) | CIDEr: 16.49 -> 17.12 | Improved |
| add event text to image input (ablation) | Image + Place · VisualCOMET val (text given) | CIDEr: 6.69 -> 17.12 | Improved |
| add EP loss (generate event and place) when text is given (ablation) | Image + Event + Place + PG · VisualCOMET val (text given) | CIDEr: 18.27 -> 15.00 | Hurt |
| add EP loss when no text is given at test time (ablation) | Image + Event + Place + PG · VisualCOMET val (no text given) | CIDEr: 8.77 -> 11.86 | Improved |
| add EP loss when no text is given at test time (ablation) | Image + Event + Place + PG · VisualCOMET val (no text given) | Acc@50: 17.35 -> 22.7 | Improved |
| add image to text-only model, human evaluation | Event + Place · VisualCOMET test, human eval (200 images) | human avg correct %: 50.1 -> 61.0 | Improved |
| add event/place text training to image-only model, human evaluation | Image + PG · VisualCOMET test, human eval (200 images) | human avg correct %: 34.4 -> 38.2 | Improved |
| use nucleus sampling instead of beam search (ablation) | Image + Event + Place + PG · VisualCOMET val | unique inferences (UI): 26.49 -> 51.99 | Improved |
Closest papers by meaning
- ConTextual: Evaluating Context-Sensitive Text-Rich Visual Reasoning in Large Multimodal Models (ICML 2024 · similarity 0.89)
- Multi-step Visual Reasoning with Visual Tokens Scaling and Verification (NeurIPS 2025 · similarity 0.88)
- VLM-R³: Region Recognition, Reasoning, and Refinement for Enhanced Multimodal Chain-of-Thought (NeurIPS 2025 · similarity 0.88)
- When Visualizing is the First Step to Reasoning: MIRA, a Benchmark for Visual Chain-of-Thought (CVPR 2026 · similarity 0.88)
- VGR: Visual Grounded Reasoning (ICLR 2026 · similarity 0.88)
- Visual Graph Arena: Evaluating Visual Conceptualization of Vision and Multimodal Large Language Models (ICML 2025 · similarity 0.87)
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.
