Research map / Image generation / Image restoration
Auto-Regressive Transformation for Image Alignment
Kanggeon Lee, Soochahn Lee, Kyoung Mu Lee
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 cross-attention layer (CAL) guidance (ablation) | ART · KBSMC | Acceptable rate (%): 51.87 -> 64.71 | Improved |
| add cross-attention layer (CAL) (ablation) | ART · GoogleMap (LR) | ACE: 0.96 -> 0.19 | Improved |
| add cross-attention layer (CAL) (ablation) | ART · ScanNet-1500 | mAUC: 45.8 -> 51.1 | Improved |
| add cross-attention layer (CAL) (ablation) | ART · FIRE | mAUC: 78.2 -> 78.5 | No clear effect |
| add cross-attention layer (CAL) (ablation) | ART · FLORI21 | mAUC: 92.3 -> 92.5 | No clear effect |
| reduce number of inference steps below default 6 (HR) or 4 (LR) | ART · HR and LR datasets | alignment accuracy | No clear effect |
| initialize transform field at higher resolution to use fewer sampling steps | ART · HR and LR datasets | alignment accuracy | Hurt |
| add multiplicative scaling parameters beyond additive translation | ART transform field · image alignment | accuracy | Improved |
| predict full affine or projective parameters instead of per-point multiplicative and additive fields | ART · image alignment | convergence | Hurt |
| use ART instead of RoMa on 3D scene pairs | ART w/ CAL · MegaDepth-1500 and ScanNet-1500 | mAUC: 74.9 vs 75.2; 51.1 vs 52.0 (RoMa) | Mixed |
Closest papers by meaning
- Jointly Aligning Millions of Images With Deep Penalised Reconstruction Congealing (ICCV 2019 · similarity 0.88)
- Dense Cross-Scale Image Alignment with Fully Spatial Correlation and Just Noticeable Difference Guidance (AAAI 2026 · similarity 0.86)
- IDEAL: In-DEpth ALignment Makes A Discrete Representation AutoEncoder (NeurIPS 2026 · similarity 0.86)
- Bringing NeRFs to the Latent Space: Inverse Graphics Autoencoder (ICLR 2025 · similarity 0.85)
- REPA Works Until It Doesn’t: Early-Stopped, Holistic Alignment Supercharges Diffusion Training (NeurIPS 2025 · similarity 0.85)
- FastJAM: a Fast Joint Alignment Model for Images (NeurIPS 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.
