Research map / Detection and segmentation / Open-set and few-shot detection
ContextFusion and Bootstrap: An Effective Approach to Improve Slot Attention-Based Object-Centric Learning
Pinzhuo Tian, Shengjie Yang, Hang Yu, Alex Chichung Kot, Bihan Wen
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 ContextFusion stage (ablation) | DINOSAUR · PASCAL | mBO^i: 44.0 -> 45.9 | Improved |
| add Bootstrap Branch on top of ContextFusion (ablation) | DINOSAUR · PASCAL | mBO^i: 45.9 -> 46.6 | Improved |
| add Bootstrap Branch on top of ContextFusion (ablation) | DINOSAUR · MOVi-C | mBO^i: 45.8 -> 47.0 | Improved |
| add ContextFusion and Bootstrap Branch (ablation) | SPOT · COCO | mBO^i: 34.7 -> 35.8 | Improved |
| feed random noise instead of foreground/background indicator into fusion (ablation) | DINOSAUR · MOVi-C | mBO^i: 42.4 -> 19.2 | Hurt |
| use our indicator instead of 2-slot DINOSAUR foreground/background information (ablation) | DINOSAUR · PASCAL | mBO^i: 41.5 -> 46.6 | Improved |
| add separation loss to pixel loss in indicator (ablation) | indicator · PASCAL | IoU: 28.8 -> 48.9 | Improved |
| add stuff-level loss to pixel and separation losses (ablation) | indicator · PASCAL | IoU: 48.9 -> 52.6 | Improved |
| add stuff-level loss to pixel loss only (no separation loss) (ablation) | indicator · PASCAL | IoU: 28.8 -> 28.4 | No clear effect |
| add cross-attention only in fusion layer instead of cross plus self attention (ablation) | DINOSAUR · MOVi-C | mBO^i: 45.0 vs 45.8 | Mixed |
| add self-attention to cross-attention in fusion layer (ablation) | SPOT · MOVi-C | mBO^i: 48.2 -> 48.8 | Improved |
| add ContextFusion and Bootstrap Branch with DINOv2 encoder (ablation) | DINOSAUR · PASCAL | mBO^i: 40.7 -> 42.9 | Improved |
Closest papers by meaning
- Grounded Object-Centric Learning (ICLR 2024 · similarity 0.91)
- Adaptive Slot Attention: Object Discovery with Dynamic Slot Number (CVPR 2024 · similarity 0.90)
- Bootstrapping Top-down Information for Self-modulating Slot Attention (NeurIPS 2024 · similarity 0.90)
- Object-Centric Pretraining via Target Encoder Bootstrapping (ICLR 2025 · similarity 0.90)
- GLASS: Guided Latent Slot Diffusion for Object-Centric Learning (CVPR 2025 · similarity 0.89)
- SPOT: Self-Training with Patch-Order Permutation for Object-Centric Learning with Autoregressive Transformers (CVPR 2024 · similarity 0.89)
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.
