Research map / Image generation / Fast diffusion sampling
Cross-Timestep: 3D Diffusion Model with Trans-temporal Memory LSTM and Adaptive Priori Decoding Strategy for Medical Segmentation
Shangqian Wu, Siyuan Shen, Yahan Li, Zhijian Huang, Ziyu Fan, Yuanpeng Zhang, YI WANG, Lei Deng
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 SC-tLSTM to APDS baseline (ablation) | Cross-Timestep · LNCTVSeg | Dice: 77.3 -> 82.1 | Improved |
| add FFT-tLSTM to APDS baseline (ablation) | Cross-Timestep · LNCTVSeg | Dice: 77.3 -> 81.8 | Improved |
| add both SC-tLSTM and FFT-tLSTM (ablation) | Cross-Timestep · LNCTVSeg | Dice: 82.1 -> 83.7 | Improved |
| add SC-tLSTM to APDS baseline (ablation) | Cross-Timestep · OASeg | Dice: 67.5 -> 71.3 | Improved |
| add the t-cell on top of LSTM variants (ablation) | tLSTM · LNCTVSeg | Dice: 82.5 -> 83.7 | Improved |
| add temporal memory (standard LSTM replaced by Conv-LSTM variant) (ablation) | tLSTM · LNCTVSeg | Dice: 79.5 -> 81.1 | Improved |
| add Linear-GRU variant versus LSTM only (ablation) | tLSTM · LNCTVSeg | Dice: 79.5 -> 79.2 | No clear effect |
| use 1000 diffusion steps instead of 300 (ablation) | Cross-Timestep · LNCTVSeg | Dice: 79.3 -> 83.7 | Improved |
| use 1000 diffusion steps instead of 500 (ablation) | Cross-Timestep · LNCTVSeg | Dice: 82.8 -> 83.7 | Improved |
| use Cross-Timestep instead of Diff-UNet | Diff-UNet · OASeg | Dice: 71.5 -> 72.8 | Improved |
| use Cross-Timestep instead of Diff-UNet (training time) | Diff-UNet · LNCTVSeg | training time (h): 29.2 -> 33.4 | Hurt |
Closest papers by meaning
- DiffusionBlend: Learning 3D Image Prior through Position-aware Diffusion Score Blending for 3D Computed Tomography Reconstruction (NeurIPS 2024 · similarity 0.88)
- TCAM-Diff: Triplane-Aware Cross-Attention Medical Diffusion Model (AAAI 2025 · similarity 0.88)
- TimeStep Master: Asymmetrical Mixture of Timestep LoRA Experts for Versatile and Efficient Diffusion Models in Vision (ICML 2025 · similarity 0.87)
- Improving Training Efficiency of Diffusion Models via Multi-Stage Framework and Tailored Multi-Decoder Architecture (CVPR 2024 · similarity 0.87)
- MedSegDiff-V2: Diffusion-Based Medical Image Segmentation with Transformer (AAAI 2024 · similarity 0.87)
- TCAQ-DM: Timestep-Channel Adaptive Quantization for Diffusion Models (AAAI 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.
