Research map / Vision-language and multimodal / Multimodal LLMs
FlashVLM: Text-Guided Visual Token Selection for Large Multimodal Models
Kaitong Cai, jusheng zhang, Sizhuo Ma, Bingqian Lu, Jian Wang, Keze Wang
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 text-guided extrinsic similarity (ablation) | FlashVLM · VQAv2 (64 tokens) | accuracy: 72.7 -> 73.6 | Improved |
| add intrinsic visual saliency (ablation) | FlashVLM · VQAv2 (64 tokens) | accuracy: 72.5 -> 73.6 | Improved |
| add diversity-preserving background reserve (ablation) | FlashVLM · VQAv2 (64 tokens) | accuracy: 72.4 -> 73.6 | Improved |
| use log-domain fusion instead of linear fusion (ablation) | FlashVLM · POPE (64 tokens) | accuracy: 80.2 -> 81.7 | Improved |
| add text sharpening (ablation) | FlashVLM · VQAv2 (64 tokens) | accuracy: 73.0 -> 73.6 | Improved |
| add query gating (ablation) | FlashVLM · VQAv2 (64 tokens) | accuracy: 72.8 -> 73.6 | Improved |
| add text guidance at 128 tokens (ablation) | FlashVLM · VQAv2 (128 tokens) | accuracy: 75.9 -> 76.4 | Improved |
| add diversity reserve at 128 tokens (ablation) | FlashVLM · GQA (128 tokens) | accuracy: 58.1 -> 58.9 | Improved |
| prune 77.8% of visual tokens (keep 128) with FlashVLM | LLaVA-1.5-7B · 10 image benchmarks | relative accuracy: 100.0% -> 100.6% | Improved |
| increase compression from 128 to 32 tokens | FlashVLM · 10 image benchmarks | relative accuracy: 100.6% -> 92.8% | Hurt |
| replace VisPruner with FlashVLM at 32 tokens | VisPruner · 10 image benchmarks | relative accuracy: 91.5% -> 92.8% | Improved |
| replace VisPruner with FlashVLM at 455 tokens | VisPruner · Video QA avg | accuracy: 48.4 -> 48.7 | 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.
- VisionZip — VisionZip: Longer is Better but Not Necessary in Vision Language Models (CVPR 2025) · better on 3 of 3 shared results
- SparseVLM — SparseVLM: Visual Token Sparsification for Efficient Vision-Language Model Inference (ICML 2025) · 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
- One Layer's Trash is Another Layer's Treasure: Adaptive Layer-wise Visual Token Selection in LVLMs (CVPR 2026 · similarity 0.93)
- Rethinking Visual Token Reduction in LVLMs Under Cross-Modal Misalignment (AAAI 2026 · similarity 0.93)
- ApET: Approximation-Error Guided Token Compression for Efficient VLMs (CVPR 2026 · similarity 0.92)
- A Stitch in Time Saves Nine: Small VLM is a Precise Guidance for Accelerating Large VLMs (CVPR 2025 · similarity 0.91)
- TopV: Compatible Token Pruning with Inference Time Optimization for Fast and Low-Memory Multimodal Vision Language Model (CVPR 2025 · similarity 0.91)
- Accelerating Multimodal Large Language Models via Dynamic Visual-Token Exit and the Empirical Findings (NeurIPS 2025 · similarity 0.91)
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.
