Research map / Image generation / Fast diffusion sampling
D3ToM: Decider-Guided Dynamic Token Merging for Accelerating Diffusion MLLMs
Shuochen Chang, Xiaofeng Zhang, Qingyang Liu, Li Niu
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 timestep-dependent linear merge schedule (D3ToM-t) over fixed ratio (ablation) | D3ToM · 7 benchmarks, retain 25% tokens | average retention: 95.23% -> 98.12% | Improved |
| add timestep-dependent schedule (ablation) | D3ToM · 7 benchmarks, retain 10% tokens | average retention: 95.31% -> 96.75% | Improved |
| add timestep-dependent schedule (ablation) | D3ToM · 7 benchmarks, retain 50% tokens | average retention: 96.85% -> 98.05% | Improved |
| use decider-guided merging instead of FastV-style pruning | D3ToM · LaViDa, retain 10% tokens | average retention: 76.34% -> 95.31% | Improved |
| use decider-guided merging instead of VisionZip | D3ToM-t · LaViDa, retain 10% tokens | average retention: 88.68% -> 96.75% | Improved |
| use D3ToM-t instead of VisionZip on inference cost | D3ToM-t · LLaVA-Bench, 10% tokens | TFLOPs: 51.36 -> 79.37 | Hurt |
| use D3ToM-t instead of FastV on inference cost | D3ToM-t · LLaVA-Bench, 10% tokens | TFLOPs: 77.14 -> 79.37 | No clear effect |
| merge visual tokens in diffusion MLLM at 10% retention | D3ToM-t · LLaVA-Bench | inference time (s): 1156.65s -> 493.88s | Improved |
| combine token merging with Prefix-DLM KV cache | D3ToM · 4 benchmarks, retain 10% tokens | average retention: 99.7% | Improved |
Closest papers by meaning
- Generative Multimodal Pretraining with Discrete Diffusion Timestep Tokens (CVPR 2025 · similarity 0.90)
- Dual Diffusion for Unified Image Generation and Understanding (CVPR 2025 · similarity 0.90)
- Boosting Multimodal Large Language Models with Visual Tokens Withdrawal for Rapid Inference (AAAI 2025 · similarity 0.90)
- I Think, Therefore I Diffuse: Enabling Multimodal In-Context Reasoning in Diffusion Models (ICML 2025 · similarity 0.90)
- ST3: Accelerating Multimodal Large Language Model by Spatial-Temporal Visual Token Trimming (AAAI 2025 · similarity 0.90)
- Thinking Diffusion: Penalize and Guide Visual-Grounded Reasoning in Diffusion Multimodal Language Models (CVPR 2026 · similarity 0.90)
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.
