Research map / Detection and segmentation / Open-set and few-shot detection
Are These the Same Apple? Comparing Images Based on Object Intrinsics
Klemen Kotar, Stephen Tian, Hong-Xing Yu, Daniel LK Yamins, Jiajun Wu
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 foreground cropping to DinoV2 (ablation) | DinoV2 · CUTE in-the-wild | mAP: 70.9 -> 79.1 | Improved |
| add foreground cropping to CLIPScore (ablation) | CLIPScore · CUTE in-the-wild | top-1: 13.3 -> 59.5 | Improved |
| add foreground cropping to CLIPScore (ablation) | CLIPScore · CUTE pose | mAP: 82.8 -> 77.7 | Hurt |
| add foreground cropping to DinoV2 (ablation) | DinoV2 · CUTE illumination | ARI: 54.0 -> 39.8 | Hurt |
| average features over foreground (Crop-Feat) instead of cropping image (Crop-Img) (ablation) | FFA · CUTE illumination | top-1: 83.0 -> 89.0 | Improved |
| Crop-Feat instead of Crop-Img (ablation) | FFA · CUTE in-the-wild | mAP: 76.7 -> 75.1 | Hurt |
| add cropping to Unicom (ablation) | Unicom · CUTE in-the-wild | top-1: 37.0 -> 66.3 | Improved |
| add cropping to DinoV1 (ablation) | DinoV1 · CUTE in-the-wild | top-1: 3.8 -> 37.8 | Improved |
| combine SpCL with FFA (Crop-Img) for vehicle Re-ID | SpCL · VeRi | top-1: 43.0 -> 44.0 | Improved |
| combine SpCL with FFA (Crop-Img) | SpCL · CityFlow | top-1: 26.7 -> 27.7 | Improved |
| combine SpCL with LPIPS | SpCL · VeRi | top-1: 43.0 -> 41.6 | Hurt |
| combine SpCL with DINOv2 + Crop | SpCL · VeRi | top-1: 43.0 -> 46.1 | Improved |
Closest papers by meaning
- DreamSim: Learning New Dimensions of Human Visual Similarity using Synthetic Data (NeurIPS 2023 · similarity 0.87)
- Efficient Discovery and Effective Evaluation of Visual Perceptual Similarity: A Benchmark and Beyond (ICCV 2023 · similarity 0.87)
- ObjectNet: A large-scale bias-controlled dataset for pushing the limits of object recognition models (NeurIPS 2019 · similarity 0.87)
- Human-like Object Grouping in Self-supervised Vision Transformers (ECCV 2026 · similarity 0.86)
- Alice Benchmarks: Connecting Real World Re-Identification with the Synthetic (ICLR 2024 · similarity 0.86)
- Multi-Part Object Representations via Graph Structures and Co-Part Discovery (NeurIPS 2026 · similarity 0.85)
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.
