Research map / Detection and segmentation / CNN features
Deep Shape Matching
Filip Radenovic, Giorgos Tolias, Ondrej Chum
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 edge filtering layer (ablation) | off-the-shelf CNN on edge maps · Flickr15k | mAP: 25.9 -> 27.9 | Improved |
| fine-tune the network with edge-map pairs (ablation) | EdgeMAC · Flickr15k | mAP: 27.9 -> 38.4 | Improved |
| add random training-query binarization (ablation) | EdgeMAC · Flickr15k | mAP: 38.4 -> 41.9 | Improved |
| add test-time mirroring (ablation) | EdgeMAC · Flickr15k | mAP: 41.9 -> 43.4 | Improved |
| add test-time multi-scale representation (ablation) | EdgeMAC · Flickr15k | mAP: 41.9 -> 45.6 | Improved |
| combine mirroring and multi-scale (ablation) | EdgeMAC · Flickr15k | mAP: 41.9 -> 46.1 | Improved |
| add diffusion re-ranking with edge-map and image similarity graph (ablation) | EdgeMAC · Flickr15k | mAP: 46.1 -> 68.9 | Improved |
| use edge maps instead of RGB input, train on photos, test on sketch | MAC descriptor · PACS (train Photo, test Sketch) | accuracy: 24.8 -> 46.3 | Improved |
| use edge maps instead of RGB input, train on photos, test on art | MAC descriptor · PACS (train Photo, test Art) | accuracy: 72.5 -> 45.4 | Hurt |
| concatenate pre-trained RGB descriptor with edge-map descriptor | EdgeMAC · PACS (mean over 4 targets) | accuracy: 65.2 -> 76.2 | Improved |
| use fine-tuned VGG instead of AlexNet as backbone (ablation) | EdgeMAC · PACS (mean over 4 targets) | accuracy: 70.9 -> 76.2 | Improved |
| add supervised descriptor whitening | EdgeMAC · Sketchy Shoes (fine-grained SBIR) | acc.@1: 40.0 -> 54.8 | Improved |
Closest papers by meaning
- Hierarchical Metric Learning and Matching for 2D and 3D Geometric Correspondences (ECCV 2018 · similarity 0.90)
- Universal Correspondence Network (NeurIPS 2016 · similarity 0.90)
- Learning 3D Keypoint Descriptors for Non-Rigid Shape Matching (ECCV 2018 · similarity 0.89)
- Image Matching from Handcrafted to Deep Features: A Survey (IJCV 2020 · similarity 0.89)
- Unsupervised Deep Multi-Shape Matching (ECCV 2022 · similarity 0.89)
- Deep Semantic Feature Matching (CVPR 2017 · similarity 0.89)
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.
