Research map / Vision-language and multimodal / Multimodal LLMs
Token Activation Map to Visually Explain Multimodal LLMs
Yi Li, Hualiang Wang, Xinpeng Ding, Haonan Wang, Xiaomeng Li
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 estimated causal inference (ECI) to CAM (ablation) | TAM · COCO Caption (Qwen2-VL-2B) | F1-IoU (%): 30.14 -> 33.84 | Improved |
| add rank Gaussian filter (RGF) to CAM (ablation) | TAM · COCO Caption (Qwen2-VL-2B) | F1-IoU (%): 30.14 -> 31.57 | Improved |
| combine ECI and RGF (ablation) | TAM · COCO Caption (Qwen2-VL-2B) | F1-IoU (%): 30.14 -> 39.1 | Improved |
| use rank Gaussian filter instead of Gaussian, median or adaptive median filter (ablation) | TAM · COCO Caption (Qwen2-VL-2B) | F1-IoU (%): 38.16 -> 39.1 | Improved |
| use ECI instead of feature surgery for context interference (ablation) | TAM · COCO Caption (Qwen2-VL-2B) | F1-IoU (%): 26.81 -> 39.1 | Improved |
| use text relevance weighting in ECI instead of plain mean of context maps (ablation) | TAM · COCO Caption (Qwen2-VL-2B) | F1-IoU (%): 35.72 -> 39.1 | Improved |
| use classifier-weight textual relevance instead of attention weights for ECI (ablation) | TAM · COCO Caption (Qwen2-VL-2B) | F1-IoU (%): 36.23 -> 39.1 | Improved |
| apply TAM on top of Layer-CAM | Layer-CAM · COCO Caption (Qwen2-VL-2B) | Func-IoU (%): +6.98 F1-IoU, -7.95 Func-IoU | Mixed |
| apply TAM to seven MLLMs instead of CAM | CAM · COCO Caption (7 MLLMs) | F1-IoU (%): +5.45 to +11.01 | Improved |
| placebo test replacing target CAM with a random earlier CAM (ablation) | TAM · COCO Caption | Obj-IoU (%): Obj-IoU reduced to 6.2% | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- Grad-CAM++ — Grad-CAM: Visual Explanations From Deep Networks via Gradient-Based Localization (ICCV 2017) · better on 3 of 3 shared results
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
- Task-Related Token Compression in Multimodal Large Language Models from an Explainability Perspective (ICLR 2026 · similarity 0.90)
- Where MLLMs Attend and What They Rely On: Explaining Autoregressive Token Generation (CVPR 2026 · similarity 0.89)
- What Do Visual Tokens Really Encode? Uncovering Sparsity and Redundancy in Multimodal Large Language Models (CVPR 2026 · similarity 0.89)
- Accelerating Multimodal Large Language Models via Dynamic Visual-Token Exit and the Empirical Findings (NeurIPS 2025 · similarity 0.89)
- TokenSwap: Benchmarking and Reducing the Modality Gap in Multimodal LLMs (NeurIPS 2026 · similarity 0.89)
- Boosting Multimodal Large Language Models with Visual Tokens Withdrawal for Rapid Inference (AAAI 2025 · similarity 0.88)
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.
