Research map / Image generation / Autoregressive and latent generation
Revisiting Self-Similarity: Structural Embedding for Image Retrieval
Seongwon Lee, Suhyeon Lee, Hongje Seong, Euntai Kim
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 SENet structural embedding to the ResNet-101 global retrieval model with classification and contrastive losses | CVNet-Global · RPar +1M, Hard | mAP: 62.2 -> 67.8 | Improved |
| Add SENet structural embedding to the reproduced ResNet-50 model | R50-DELG† · ROxf +1M, Hard | mAP: 44.8 -> 48.7 | Improved |
| Add the self-similarity encoder without zero-initialized BN or the feed-forward layer (ablation) | ResNet-50 baseline · RPar +1M, Hard | mAP: 57.7 -> 56.3 | Hurt |
| Add zero-initialized BN before summing the structural and original features (ablation) | SENet with SSE · RPar +1M, Hard | mAP: 56.3 -> 58.3 | Improved |
| Add the pixel-wise feed-forward layer to fuse the features (ablation) | SENet with SSE and BNinit · RPar +1M, Hard | mAP: 58.3 -> 59.9 | Improved |
| Extract structural information from conv5 rather than conv4 (ablation) | SENet · RPar +1M, Hard | mAP: 57.2 -> 59.9 | Improved |
| Set self-similarity region size P to 5 instead of P = 7 (ablation) | SENet · RPar +1M, Hard | mAP: 59.9 -> 58.5 | Hurt |
| Set self-similarity region size P to 9 instead of P = 7 (ablation) | SENet · RPar +1M, Hard | mAP: 59.9 -> 59.9 | No clear effect |
Closest papers by meaning
- Scene Graph Embeddings Using Relative Similarity Supervision (AAAI 2021 · similarity 0.89)
- Beyond Instance-Level Image Retrieval: Leveraging Captions to Learn a Global Visual Representation for Semantic Retrieval (CVPR 2017 · similarity 0.88)
- Guided Similarity Separation for Image Retrieval (NeurIPS 2019 · similarity 0.88)
- Deep Image Retrieval: Learning Global Representations for Image Search (ECCV 2016 · similarity 0.88)
- Exploiting Distribution Constraints for Scalable and Efficient Image Retrieval (ICLR 2025 · similarity 0.86)
- Instance-Level Composed Image Retrieval (NeurIPS 2025 · similarity 0.86)
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.
