Research map / Detection and segmentation / Open-set and few-shot detection
Object Discovery and Representation Networks
Olivier J. Hénaff, Skanda Koppula, Evan Shelhamer, Daniel Zoran, Andrew Jaegle, Andrew Zisserman, João Carreira, Relja Arandjelović
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 K=16 instead of K=8 segments in k-means (ablation) | Odin · COCO object discovery (ResNet-50) | ABO instance: 38.3 -> 43.0 | Improved |
| use K=32 instead of K=16 segments (ablation) | Odin · COCO object discovery (ResNet-50) | ABO instance: 43.0 -> 43.1 | No clear effect |
| use continuous discovery-network updates (1e-3) instead of every 100 epochs (ablation) | Odin · COCO Mask-RCNN 1x | AP bbox: 42.6 -> 42.9 | Mixed |
| continuous updates (1e-3) versus discrete updates for object discovery (ablation) | Odin · COCO object discovery (ResNet-50) | ABO instance: 43.0 -> 41.3 | Mixed |
| replace hand-crafted segmentation (DetCon) with learned object discovery | DetCon_B · COCO object discovery (ResNet-50) | ABO instance: 34.1 -> 41.5 | Improved |
| replace hand-crafted segmentation with learned discovery for COCO transfer | DetCon_B · COCO Mask-RCNN 1x | AP bbox: 42.7 -> 42.9 | Improved |
| replace hand-crafted segmentation with learned discovery | DetCon_B · COCO FCOS-style detector 30 ep (ResNet-50) | AP bbox: 45.4 -> 45.6 | No clear effect |
| pretrain with Odin instead of DINO | DINO · COCO FCOS-style detector 30 ep (ResNet-50) | AP bbox: 44.3 -> 45.6 | Improved |
| pretrain Swin-S with Odin instead of supervised | Supervised · COCO FCOS-style detector 30 ep (Swin-S) | AP bbox: 48.3 -> 50.4 | Improved |
| keep FPN when evaluating object discovery | Odin · COCO object discovery (ResNet-50) | ABO instance: 41.5 -> 43.0 | Improved |
| train a ViT-B backbone with Odin instead of DINO | DINO · COCO object discovery (ViT-B) | ABO instance: 42.7 -> 45.9 | Improved |
| fine-tune the target network instead of the online network (ablation) | Odin · COCO Mask-RCNN 1x | AP bbox: 42.9 -> 42.7 | No clear effect |
Closest papers by meaning
- Unsupervised Semantic Segmentation with Self-supervised Object-centric Representations (ICLR 2023 · similarity 0.91)
- Self-Supervised Learning of Object Parts for Semantic Segmentation (CVPR 2022 · similarity 0.91)
- Unsupervised Object-Level Representation Learning from Scene Images (NeurIPS 2021 · similarity 0.89)
- UniT: Unified Knowledge Transfer for Any-Shot Object Detection and Segmentation (CVPR 2021 · similarity 0.89)
- Self-Supervised Learning of Intertwined Content and Positional Features for Object Detection (ICML 2025 · similarity 0.88)
- DIOD: Self-Distillation Meets Object Discovery (CVPR 2024 · similarity 0.87)
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.
