Research map / Label-efficient and robust learning / Recognition and long-tail
SUBIC: A Supervised, Structured Binary Code for Image Search
Himalaya Jain, Joaquin Zepeda, Patrick Perez, Remi Gribonval
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 supervised block-softmax encoding instead of unsupervised LSQ quantization | VGG-128 features · VOC2007 cross-domain retrieval | mAP: 0.4993 -> 0.5600 | Improved |
| add supervised encoding instead of unsupervised LSQ | VGG-128 features · ImageNet val retrieval | mAP: 0.1882 -> 0.2543 | Improved |
| train FC1 as well (3-layer instead of 2-layer) (ablation) | SuBiC · VOC2007 cross-domain retrieval | mAP: 0.5600 -> 0.5588 | No clear effect |
| train FC1 as well (3-layer instead of 2-layer) (ablation) | SuBiC · Caltech-101 cross-domain retrieval | mAP: 0.3923 -> 0.4033 | Improved |
| train FC1 as well (3-layer instead of 2-layer) (ablation) | SuBiC · ImageNet val retrieval | mAP: 0.2543 -> 0.2810 | Improved |
| increase bit rate from 12 to 48 bits (ablation) | SuBiC · Cifar-10 category retrieval | mAP: 0.6349 -> 0.6863 | Improved |
| use SuBiC instead of the supervised hashing method DSH | 64-bit codes · Oxford5K instance retrieval | mAP: 0.2108 -> 0.2626 | Improved |
| use SuBiC 64-bit code instead of PQ for compressed classification | VGG-128 · ImageNet val classification | top-1 accuracy: 39.88 -> 47.77 | Improved |
| binarize the soft code to a 64-bit binary code (ablation) | SuBiC · ImageNet val classification | top-1 accuracy: 50.07 -> 47.77 | Hurt |
| binarize the soft code to a 64-bit binary code (ablation) | SuBiC · VOC2007 classification | mAP: 70.20 -> 67.86 | Hurt |
| set entropy weight ratio gamma/mu to 1 instead of other ratios (ablation) | SuBiC · category retrieval | mAP | Improved |
| increase mean-entropy weight gamma relative to mu (ablation) | SuBiC · ImageNet val and VOC2007 code statistics | one-hot closeness | Improved |
Later papers that compared with this method
2 later accepted papers report SUBIC as a baseline in their results tables; 2 of them report a better number on most of the results they share with it.
- End-To-End Supervised Product Quantization for Image Search and Retrieval (CVPR 2019 · better on 5 of 5 shared results)
- Product Quantization Network for Fast Image Retrieval (ECCV 2018 · better on 4 of 4 shared results)
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Deep Supervised Hashing for Fast Image Retrieval (CVPR 2016 · similarity 0.89)
- Deep Variational and Structural Hashing (TPAMI 2018 · similarity 0.89)
- Self-Supervised Product Quantization for Deep Unsupervised Image Retrieval (ICCV 2021 · similarity 0.89)
- Unsupervised Neural Quantization for Compressed-Domain Similarity Search (ICCV 2019 · similarity 0.89)
- Deep Quantization Network for Efficient Image Retrieval (AAAI 2016 · similarity 0.88)
- Fast Training of Triplet-Based Deep Binary Embedding Networks (CVPR 2016 · similarity 0.88)
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.
