Research map / Label-efficient and robust learning / Recognition and long-tail
Simultaneous Feature Aggregating and Hashing for Large-Scale Image Search
Check 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 |
|---|---|---|---|
| Jointly learn feature aggregation and hashing rather than aggregate with GMP before hashing | SAH versus RBA · Holidays and Oxford5k, SIFT features | mAP: 2%-3% mAP | Helped |
| Jointly learn feature aggregation and hashing rather than aggregate with GMP before hashing | SAH versus other hashing methods · Holidays, Holidays+Flickr100k, and Oxford5k, SIFT features | mAP | Helped |
| Use convolutional feature maps rather than SIFT features | SAH · Holidays+Flickr100k | mAP: about 5%, 10%, 14% for L=16,24 and 32, respectively | Helped |
| Use convolutional feature maps rather than SIFT features | SAH · Holidays and Holidays+Flickr100k at L=8 | mAP | No clear effect |
| Use SAH with convolutional features instead of BA | SAH versus BA · Holidays, Holidays+Flickr100k, and Oxford5k at L=32 | mAP: 2%-3% mAP | Helped |
| Use convolutional rather than fully connected features with SAH | SAH · Holidays and Holidays+Flickr100k at L=8 | mAP | No clear effect |
| Jointly learn aggregation and hashing using VGG convolutional features | SAH versus BA · CIFAR10 test set (top 1,000) | mAP: 22.3%, 20.7%, 19.6% at L=16, 32 and 64, respectively | Helped |
| Replace the hard binary optimization with the relaxed binary quantization objective and alternating updates | RBA versus BA · CIFAR10, MNIST, and SIFT1M | mAP | No clear effect |
| Use RBA instead of BA | RBA versus BA · CIFAR10 and SIFT1M | training time: more than ten times faster | Helped |
| Use VGG fully connected features rather than GIST features | ITQ-CNN, KMH-CNN, and SPH-CNN · CIFAR10 test set (top 1,000) | mAP | Helped |
| Use SAH rather than DeepBit | SAH versus DeepBit · CIFAR10 test set (top 1,000) | mAP: 22.3%, 20.7%, 19.6% at L=16, 32 and 64 | Helped |
| Learn the aggregator and hash function jointly rather than using GMP aggregation before hashing | SAH versus ITQ, BA, and RBA · Oxford5k | reconstruction error | Helped |
Closest papers by meaning
- Deep Supervised Hashing for Fast Image Retrieval (CVPR 2016 · similarity 0.90)
- Learning Attribute-Aware Hash Codes for Fine-Grained Image Retrieval via Query Optimization (ICML 2025 · similarity 0.89)
- A$^2$-Net: Learning Attribute-Aware Hash Codes for Large-Scale Fine-Grained Image Retrieval (NeurIPS 2021 · similarity 0.89)
- Learning Multifunctional Binary Codes for Both Category and Attribute Oriented Retrieval Tasks (CVPR 2017 · similarity 0.89)
- Pairwise-Label-Based Deep Incremental Hashing with Simultaneous Code Expansion (AAAI 2024 · similarity 0.85)
- An Asymmetric Augmented Self-Supervised Learning Method for Unsupervised Fine-Grained Image Hashing (CVPR 2024 · 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.