Research map / Label-efficient and robust learning / Recognition and long-tail
Unsupervised Part-Based Weighting Aggregation of Deep Convolutional Features for Image Retrieval
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| select part detectors by channel variance rather than randomly (ablation) | PWA · Oxford5k | mAP: 0.775±0.006 -> 0.790 | Helped |
| select the 25 highest-variance detectors instead of all 512 (ablation) | PWA · Oxford5k | mAP: 78.5 -> 79.1 | Helped |
| select the 25 highest-variance detectors instead of all 512 (ablation) | PWA · Paris6k | mAP: 85.4 -> 86.1 | Helped |
| increase PCA output dimensionality from 128 to 4096 (ablation) | PWA · Oxford5k | mAP: 64.5 -> 79.1 | Helped |
| increase PCA output dimensionality from 128 to 4096 (ablation) | PWA · Paris6k | mAP: 76.9 -> 86.1 | Helped |
| use semantic part-based weighting and concatenation rather than standard unsupervised aggregation | PWA · Oxford5k | mAP: 70.8 -> 79.1 | Helped |
| use semantic part-based weighting and concatenation rather than standard unsupervised aggregation | PWA · Paris6k | mAP: 83.0 -> 86.1; 79.2 -> 86.1 | Mixed |
| use PWA with fine-tuned convolutional features instead of DeepRepresentation | PWA · Oxford5k | mAP: 86.1 -> 87.8 | Helped |
| use PWA with fine-tuned convolutional features instead of DeepRepresentation | PWA · Paris6k | mAP: 94.5 -> 94.9 | Helped |
| use PWA with fine-tuned convolutional features instead of DeepRepresentation | PWA · Oxford105k | mAP: 82.8 -> 82.8 | No clear effect |
| use PWA with fine-tuned convolutional features instead of DeepRepresentation | PWA · Paris106k | mAP: 90.6 -> 91.0 | Helped |
Closest papers by meaning
- Unsupervised Part Learning for Visual Recognition (CVPR 2017 · similarity 0.89)
- Learning Token-Based Representation for Image Retrieval (AAAI 2022 · similarity 0.85)
- Weakly Supervised Complementary Parts Models for Fine-Grained Image Classification From the Bottom Up (CVPR 2019 · similarity 0.85)
- SPDA-CNN: Unifying Semantic Part Detection and Abstraction for Fine-Grained Recognition (CVPR 2016 · similarity 0.84)
- Instance-Level Composed Image Retrieval (NeurIPS 2025 · similarity 0.84)
- Unsupervised Deep Learning via Affinity Diffusion (AAAI 2020 · similarity 0.84)
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.