Research map / Vision-language and multimodal / Cross-modal learning
UniStitch: Unifying Semantic and Geometric Features for Image Stitching
Yuan Mei, Lang Nie, Kang Liao, Yunqiu Xu, Chunyu Lin, Bin Xiao
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 semantic image branch to point-only model (ablation) | UniStitch (point-only) · UDIS-D | mPSNR: 24.22 -> 24.63 | Improved |
| add geometric point branch to image-only model (ablation) | UniStitch (image-only) · Classic datasets | mPSNR: 17.91 -> 18.02 | Improved |
| add AMoE fusion of point and image features (ablation) | UniStitch · Classic datasets | mPSNR: 18.02 -> 18.56 | Improved |
| add AMoE fusion of point and image features (ablation) | UniStitch · UDIS-D | mPSNR: 24.63 -> 24.94 | Improved |
| add Modality Robustifier (MR) on top of AMoE (ablation) | UniStitch · Classic datasets | mPSNR: 18.56 -> 18.80 | Improved |
| add Modality Robustifier (MR) on top of AMoE (ablation) | UniStitch · UDIS-D | mPSNR: 24.94 -> 25.07 | Improved |
| use AMoE instead of simple Add fusion (ablation) | UniStitch · Classic datasets | mPSNR: 17.22 -> 18.56 | Improved |
| use AMoE instead of CSPFusion (ablation) | UniStitch · Classic datasets | mPSNR: 18.52 -> 18.56 | Improved |
| use matched keypoints instead of raw keypoints for training (ablation) | UniStitch · Classic datasets | mPSNR: 17.90 -> 18.48 | Improved |
| randomize keypoint order at test time (ablation) | UniStitch · UDIS-D | mPSNR: 24.78 -> 24.76 | No clear effect |
| use PointTransformer v3 instead of PointNeXt as point encoder (ablation) | UniStitch (point-only) · UDIS-D | mPSNR: 24.22 -> 23.96 | Hurt |
| use FFD-based TPS instead of vanilla TPS (ablation) | UniStitch · 1329x2000 resolution | peak GPU memory (GB): 13.99 -> 8.57 | Improved |
Closest papers by meaning
- Parallax-Tolerant Unsupervised Deep Image Stitching (ICCV 2023 · similarity 0.89)
- PixelStitch: Structure-Preserving Pixel-Wise Bidirectional Warps for Unsupervised Image Stitching (ICCV 2025 · similarity 0.88)
- Reconstructing the Image Stitching Pipeline: Integrating Fusion and Rectangling into a Unified Inpainting Model (NeurIPS 2024 · similarity 0.88)
- Object-level Geometric Structure Preserving for Natural Image Stitching (AAAI 2025 · similarity 0.87)
- Depth-Supervised Fusion Network for Seamless-Free Image Stitching (NeurIPS 2025 · similarity 0.87)
- Towards Robust Image Stitching: An Adaptive Resistance Learning against Compatible Attacks (AAAI 2024 · 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.
