Research map / Architectures and efficiency / Attention and transformers
SpaceJAM: a Lightweight and Regularization-free Method for Fast Joint Alignment of Images
Nir Barel, Ron A Shapira Weber, Nir Mualem, Shahaf E Finder, Oren Freifeld
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 autoencoder pre-training before joint alignment (ablation) | SpaceJAM · CUB-200 (3 categories) | PCK@0.1: 54.9 -> 73.3 | Improved |
| add autoencoder pre-training before joint alignment (ablation) | SpaceJAM · SPair-71k (3 categories) | PCK@0.1: 56.6 -> 58.1 | Improved |
| predict a cascade of 5 small transformations instead of a single large one (ablation) | SpaceJAM · SPair-71k (3 categories) | PCK@0.1: 43.5 -> 58.1 | Improved |
| use cascade of 5 instead of 3 transformations (ablation) | SpaceJAM · CUB-200 (3 categories) | PCK@0.1: 72.6 -> 73.3 | Improved |
| use inverse-compositional loss instead of alignment to an atlas (ablation) | SpaceJAM · CUB-200 (3 categories) | PCK@0.1: 58.5 -> 73.3 | Improved |
| use Lie group parameterization (ablation) | SpaceJAM · CUB-200 (3 categories) | PCK@0.1: 65.6 -> 73.3 | Improved |
| add Lie-algebraic curriculum learning (ablation) | SpaceJAM · CUB-200 (3 categories) | PCK@0.1: 70.8 -> 73.3 | Improved |
| add Lie-algebraic curriculum learning (ablation) | SpaceJAM · SPair-71k (3 categories) | PCK@0.1: 57.2 -> 58.1 | Improved |
| use DINOv2 ViT-L features instead of DINOv1 ViT-S | SpaceJAM · SPair-71k (all) | PCK@0.10: 37.8 -> 45.7 | Improved |
| use lightweight SpaceJAM (16K params, ~6 min) instead of ASIC (7.9M params) | joint alignment · SPair-71k (all) | PCK@0.10: 36.9 -> 37.8 | Improved |
| use SpaceJAM instead of ASIC on CUB-200 3 categories | joint alignment · CUB-200 (3 categories) | PCK@0.10: 75.9 -> 69.6 | Hurt |
Closest papers by meaning
- FastJAM: a Fast Joint Alignment Model for Images (NeurIPS 2025 · similarity 0.93)
- RUBIK: A Structured Benchmark for Image Matching across Geometric Challenges (CVPR 2025 · similarity 0.84)
- Topology-Aware Representation Alignment for Semi-Supervised Vision-Language Learning (NeurIPS 2026 · similarity 0.84)
- Locality Alignment Improves Vision-Language Models (ICLR 2025 · similarity 0.84)
- Assessing and Learning Alignment of Unimodal Vision and Language Models (CVPR 2025 · similarity 0.83)
- Optimizing Distributional Geometry Alignment with Optimal Transport for Generative Dataset Distillation (NeurIPS 2025 · similarity 0.83)
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.
