Research map / Detection and segmentation / Open-set and few-shot detection
Object-centric Learning with Cyclic Walks between Parts and Whole
Ziyu Wang, Mike Zheng Shou, Mengmi Zhang
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 |
|---|---|---|---|
| use only P-W-P walks instead of walks in both directions (ablation) | Cyclic walks · Pascal VOC 2012 / COCO 2017 | ARI-FG: 29.6 -> 27.1 (VOC); 39.7 -> 34.8 (COCO) | Hurt |
| use only W-P-W walks instead of both directions (ablation) | Cyclic walks · Pascal VOC 2012 / COCO 2017 | ARI-FG: 29.6 -> 28.4 (VOC); 39.7 -> 36.4 (COCO) | Hurt |
| lower walk temperature from default 0.1 to 0.07 (ablation) | Cyclic walks · Pascal VOC 2012 | ARI-FG: 29.6 -> 23.5 | Hurt |
| raise walk temperature to 1 (ablation) | Cyclic walks · Pascal VOC 2012 | ARI-FG: 29.6 -> 21.9 | Hurt |
| increase the number of slots (to 7 on VOC, to 13 on COCO) (ablation) | Cyclic walks · Pascal VOC 2012 / COCO 2017 | ARI-FG: 29.6 -> 24.3 (VOC); 39.7 -> 35.3 (COCO) | Hurt |
| remove the similarity threshold in P-W-P walks (-inf) (ablation) | Cyclic walks · Pascal VOC 2012 / COCO 2017 | ARI-FG: 29.6 -> 22.5 (VOC); 39.7 -> 29.8 (COCO) | Hurt |
| raise similarity threshold to 1 (ablation) | Cyclic walks · Pascal VOC 2012 / COCO 2017 | ARI-FG: 29.6 -> 26.2 (VOC); 39.7 -> 31.8 (COCO) | Hurt |
| replace DINO with MAE feature extractor (ablation) | Cyclic walks · Pascal VOC 2012 / COCO 2017 | ARI-FG: 29.6 -> 26.8 (VOC); 39.7 -> 34.5 (COCO) | Hurt |
| replace DINO with MoCo-v3 or MSN feature extractor (ablation) | Cyclic walks · Pascal VOC 2012 | ARI-FG: 29.6 -> 29.3 (MoCo-v3); 29.6 -> 29.0 (MSN) | No clear effect |
| add a transformer decoder with pixel reconstruction loss (ablation) | Cyclic walks · Pascal VOC 2012 | ARI-FG: 29.6 -> 29.9 | No clear effect |
| fine-tune the DINO feature extractor jointly instead of freezing it (ablation) | Cyclic walks · Pascal VOC 2012 | ARI-FG: 29.6 -> 12.2 | Hurt |
| use a smaller learning rate for the feature extractor when fine-tuning (ablation) | Cyclic walks with fine-tuned DINO · Pascal VOC 2012 | ARI-FG: 12.2 -> 22.4 | Improved |
Closest papers by meaning
- Cycle Consistency Driven Object Discovery (ICLR 2024 · similarity 0.90)
- GLASS: Guided Latent Slot Diffusion for Object-Centric Learning (CVPR 2025 · similarity 0.89)
- MetaSlot: Break Through the Fixed Number of Slots in Object-Centric Learning (NeurIPS 2025 · similarity 0.89)
- Bootstrapping Top-down Information for Self-modulating Slot Attention (NeurIPS 2024 · similarity 0.89)
- Adaptive Slot Attention: Object Discovery with Dynamic Slot Number (CVPR 2024 · similarity 0.89)
- Grounded Object-Centric Learning (ICLR 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.
