Research map / Vision-language and multimodal / Cross-modal learning
When Lines Meet Textures: Spatial-Frequency Aligned Diffusion Features for Cross-Sparsity Correspondence
Mingrui Zhu, Fengzhi Wang, Xin Wei, Jun Wang, Nannan Wang, Xinbo Gao
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 LoFFA (low-frequency aggregation) to Unified CleanDIFT + Conv (ablation) | SFA-DIFT · PSC6K+MS-PSC6K averaged | PCK@5: 68.78 -> 69.83 | Improved |
| add AdaIN distribution alignment in LoFFA (ablation) | SFA-DIFT · PSC6K+MS-PSC6K averaged | PCK@5: 68.64 -> 69.83 | Improved |
| use LoFE instead of plain conv (ablation) | SFA-DIFT · PSC6K+MS-PSC6K averaged | PCK@5: 69.01 -> 69.83 | Improved |
| use DWT/IDWT instead of conv (ablation) | SFA-DIFT · PSC6K+MS-PSC6K averaged | PCK@5: 66.98 -> 69.83 | Improved |
| use two-level wavelet decomposition rather than single (ablation) | SFA-DIFT · PSC6K+MS-PSC6K averaged | PCK@5: 68.45 -> 69.83 | Improved |
| LoRA fine-tune CleanDIFT on sketches (Unified CleanDIFT), zero-shot (ablation) | CleanDIFT · PSC6K+MS-PSC6K averaged | PCK@5: 57.27 -> 58.69 | Improved |
| add DINOv2 features to Unified CleanDIFT, zero-shot (ablation) | Unified CleanDIFT · PSC6K+MS-PSC6K averaged | PCK@5: 58.69 -> 56.76 | Hurt |
| full SFA-DIFT vs CleanDIFT+DINO baseline | CleanDIFT+DINO · PSC6K | PCK@5: 70.25 -> 72.94 | Improved |
| full SFA-DIFT vs CleanDIFT+DINO on stylized data | CleanDIFT+DINO · MS-PSC6K Abstract | PCK@5: 66.02 -> 70.43 | Improved |
| add LoFFA on top of Unified CleanDIFT for style robustness (ablation) | Unified CleanDIFT · MS-PSC6K robustness ratio | RR@1: 0.99 -> 0.87 | Hurt |
| fine-tune baselines on MS-PSC6K | Self-Sup / SketchFusion · MS-PSC6K Abstract | PCK@5: Self-Sup 51.96 -> 45.23 | Hurt |
Closest papers by meaning
- Self-Supervised Spatial Correspondence Across Modalities (CVPR 2025 · similarity 0.87)
- Balancing Frequencies and Pixels in Flow Matching (NeurIPS 2026 · similarity 0.87)
- Unleashing Diffusion Transformers for Visual Correspondence by Modulating Massive Activations (NeurIPS 2025 · similarity 0.85)
- MDFL: Multi-Domain Diffusion-Driven Feature Learning (AAAI 2024 · similarity 0.85)
- Color Alignment in Diffusion (CVPR 2025 · similarity 0.85)
- FAM Diffusion: Frequency and Attention Modulation for High-Resolution Image Generation with Stable Diffusion (CVPR 2025 · similarity 0.85)
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.
