Research map / Label-efficient and robust learning / Recognition and long-tail
DeepCD: Learning Deep Complementary Descriptors for Patch Representations
Tsun-Yi Yang, Jo-Han Hsu, Yen-Yu Lin, Yung-Yu Chuang
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 a 32-byte binary complementary descriptor stream (splitting model) to PNNet-128 | PNNet 128-d · Brown dataset real-valued descriptors (mean of 6 train/test combos) | FPR95 (%): 7.26 -> 6.17 | Improved |
| use a separate 2-stream complementary network instead of the splitting model (ablation) | DeepCD · Brown dataset real-valued descriptors (mean of 6 train/test combos) | FPR95 (%): 6.17 -> 5.69 | Improved |
| add the DDM layer to DeepCD 2-stream (ablation) | DeepCD 2-stream · Brown dataset real-valued descriptors (mean of 6 train/test combos) | FPR95 (%): 5.69 -> 5.48 | Improved |
| double PNNet descriptor dimensionality to 256 instead of adding a complementary descriptor (ablation) | PNNet · Brown dataset real-valued descriptors (mean of 6 train/test combos) | FPR95 (%): 7.26 -> 6.98 | Improved |
| combine two PNNets with compact bilinear pooling (ablation) | PNNet · Brown dataset real-valued descriptors (mean of 6 train/test combos) | FPR95 (%) | No clear effect |
| add the DDM layer to DeepCD 2-stream binary descriptors (128+64 bits) (ablation) | DeepCD 2-stream binary · Brown dataset binary descriptors (mean of 6 train/test combos) | FPR95 (%): 13.41 -> 13.34 | Improved |
| add the DDM layer to DeepCD 2-stream binary descriptors (512+256 bits) (ablation) | DeepCD 2-stream binary · Brown dataset binary descriptors (mean of 6 train/test combos) | FPR95 (%): 7.59 -> 7.55 | Improved |
| use a 2-stream instead of splitting model for binary descriptors (128+64 bits) (ablation) | DeepCD binary · Brown dataset binary descriptors (mean of 6 train/test combos) | FPR95 (%): 17.83 -> 13.41 | Improved |
| add the DDM layer to DeepCD 2-stream (ablation) | DeepCD 2-stream · Strecha wide baseline matching (mean of fountain and herzjesu) | mAP: 0.4518 -> 0.4588 | Improved |
| add complementary descriptor to PNNet for wide baseline matching | PNNet · Strecha wide baseline matching (mean of fountain and herzjesu) | mAP: 0.3948 -> 0.4518 | Improved |
| add the DDM layer to DeepCD 2-stream (ablation) | DeepCD 2-stream · Oxford descriptor benchmark | mAP: 0.5726 -> 0.5773 | Improved |
| add complementary descriptor to PNNet on Oxford | PNNet · Oxford descriptor benchmark | mAP: 0.5541 -> 0.5726 | Improved |
Closest papers by meaning
- L2-Net: Deep Learning of Discriminative Patch Descriptor in Euclidean Space (CVPR 2017 · similarity 0.87)
- Convolutional Patch Representations for Image Retrieval: An Unsupervised Approach (IJCV 2016 · similarity 0.86)
- Explicit Spatial Encoding for Deep Local Descriptors (CVPR 2019 · similarity 0.85)
- Descriptor Distillation: A Teacher-Student-Regularized Framework for Learning Local Descriptors (IJCV 2024 · similarity 0.85)
- Efficient Learning of CNNs using Patch Based Features (ICML 2022 · similarity 0.84)
- A Unified, Resilient, and Explainable Adversarial Patch Detector (CVPR 2025 · similarity 0.82)
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.
