Research map / Image generation / Autoregressive and latent generation
Improved Object-Centric Diffusion Learning with Registers and Contrastive Alignment
Bac Nguyen, Yuhta Takida, Naoki Murata, Chieh-Hsin Lai, Toshimitsu Uesaka, Stefano Ermon, Yuki Mitsufuji
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 register slots (ablation) | SD baseline + CA · COCO | FG-ARI: 36.99 -> 45.95 | Improved |
| add register slots (ablation) | SD baseline · COCO | FG-ARI: 20.99 -> 23.64 | Improved |
| finetune cross-attention key/value/output projections (ablation) | SD baseline · COCO | FG-ARI: 20.99 -> 36.99 | Improved |
| add contrastive alignment loss (ablation) | CODA (Reg+CA) · COCO | FG-ARI: 45.95 -> 47.54 | Improved |
| add contrastive alignment loss (ablation) | CODA (Reg+CA) · COCO compositional generation | FID: 29.34 -> 31.03 | Hurt |
| use a single learnable register slot instead of none (ablation) | CODA · VOC | FG-ARI: 15.44 -> 30.39 | Improved |
| increase learnable registers from 1 to 77 (ablation) | CODA · VOC | FG-ARI: 30.39 -> 30.21 | No clear effect |
| use frozen padding registers instead of learnable (ablation) | CODA · VOC | FG-ARI: 30.21 -> 32.23 | Improved |
| unfreeze diffusion decoder under contrastive loss (ablation) | CODA · VOC | all metrics | Hurt |
| use CODA over SlotAdapt | CODA · COCO | FG-ARI: 41.4 -> 47.54 | Improved |
| use CODA over best baseline | CODA · MOVi-C/E 3D bounding box | MSE: 1.44 -> 2.11 | Hurt |
| use CFG scale 2.0 (ablation) | CODA · COCO generation | FID/KID | Improved |
Closest papers by meaning
- Slot-Guided Adaptation of Pre-trained Diffusion Models for Object-Centric Learning and Compositional Generation (ICLR 2025 · similarity 0.92)
- GLASS: Guided Latent Slot Diffusion for Object-Centric Learning (CVPR 2025 · similarity 0.92)
- Adaptive Slot Attention: Object Discovery with Dynamic Slot Number (CVPR 2024 · similarity 0.89)
- Grounded Object-Centric Learning (ICLR 2024 · similarity 0.89)
- MetaSlot: Break Through the Fixed Number of Slots in Object-Centric Learning (NeurIPS 2025 · similarity 0.89)
- Identifiable Object-Centric Representation Learning via Probabilistic Slot Attention (NeurIPS 2024 · similarity 0.88)
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.
