Research map / Label-efficient and robust learning / Recognition and long-tail
Fast Training of Triplet-Based Deep Binary Embedding Networks
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 |
|---|---|---|---|
| Use triplet-based retrieval supervision instead of pairwise supervision | Hashing methods · NUS-WIDE setting-2 and ranking-based retrieval evaluations | similarity precision | Helped |
| Interleave code inference with hash-function learning, updating code groups after each stage | Proposed hashing method · IJB-A | face search accuracy | Helped |
| Reduce the group length from 32 bits to 8 bits (ablation) | Proposed hashing method · IJB-A | CMC Rank-1: 0.818 \pm 0.023 -> 0.831\pm 0.020 | Helped |
| Reduce the group length from 64 bits to 8 bits (ablation) | Proposed hashing method · IJB-A | CMC Rank-1: 0.793 \pm 0.024 -> 0.831\pm 0.020 | Helped |
| Reduce the group length from 128 bits to 8 bits (ablation) | Proposed hashing method · IJB-A | CMC Rank-1: 0.778 \pm 0.023 -> 0.831\pm 0.020 | Helped |
| Reduce the group length from 128 bits to 8 bits (ablation) | Proposed hashing method · IJB-A | FNIR @ FPIR 0.01: {\rm{0.645}} \pm 0.058 -> 0.598 \pm 0.048 | Helped |
| Use the two-step code-inference and CNN hash-function learning framework rather than direct triplet CNN training | Proposed hashing method · MIT Indoor | training time (hours): 186 -> 18 | Helped |
| Use the two-step code-inference and CNN hash-function learning framework rather than direct triplet CNN training | Proposed hashing method · CIFAR-10 | training time (hours): 174 -> 15 | Helped |
| Use the two-step code-inference and CNN hash-function learning framework rather than direct triplet CNN training | Proposed hashing method · NUS-WIDE setting-1 | training time (hours): 365 -> 32 | Helped |
| Use triplet ranking loss rather than pairwise similarity supervision | Proposed hashing method · Retrieval benchmarks | MAP | Helped |
Closest papers by meaning
- Deep Supervised Hashing for Fast Image Retrieval (CVPR 2016 · similarity 0.90)
- Efficient end-to-end learning for quantizable representations (ICML 2018 · similarity 0.89)
- RankMI: A Mutual Information Maximizing Ranking Loss (CVPR 2020 · similarity 0.87)
- Binary Generative Adversarial Networks for Image Retrieval (AAAI 2018 · similarity 0.87)
- LLC: Accurate, Multi-purpose Learnt Low-dimensional Binary Codes (NeurIPS 2021 · similarity 0.87)
- Pairwise-Label-Based Deep Incremental Hashing with Simultaneous Code Expansion (AAAI 2024 · 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.