Research map / Label-efficient and robust learning / Recognition and long-tail
Do Different Deep Metric Learning Losses Lead to Similar Learned Features?
Konstantin Kobs, Michael Steininger, Andrzej Dulny, Andreas Hotho
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 |
|---|---|---|---|
| compare loss pairs of different type (ranking vs classification) instead of same type | DML saliency maps · SOP | saliency correlation (%): 62 -> 46 | Hurt |
| compare ranking with classification losses instead of same type | DML saliency maps · Cars196 | saliency correlation (%): 86 -> 85 | No clear effect |
| compare ranking with classification losses instead of same type | DML saliency maps · CUB200 | saliency correlation (%): 90 -> 90 | No clear effect |
| compare trained model against untrained None model | DML saliency maps · Cars196 | saliency correlation (%): 86 -> 75 | Hurt |
| train with DML loss rather than ImageNet init only (None) | car model clustering · Rendered cars: car model | NR-Prec: 50.61 -> 57.91 | Improved |
| train with DML loss rather than ImageNet init only (None) | background color value clustering · Rendered cars: background color value | NR-Prec: 6.80 -> 1.62 | Improved |
| use classification instead of ranking loss | car color hue influence · Rendered cars: car color hue | NR-Prec: 3.25 -> 4.07 | Hurt |
| use classification instead of ranking loss | background color saturation influence · Rendered cars: background color saturation | NR-Prec: 1.48 -> 2.22 | Hurt |
| use classification instead of ranking loss | sun elevation influence · Rendered cars: sun elevation | NR-Prec: 8.29 -> 8.94 | Hurt |
| use classification instead of ranking loss | car rotation influence · Rendered cars: car rotation | NR-Prec: 39.07 -> 37.84 | Improved |
| use classification instead of ranking loss | camera height influence · Rendered cars: camera height | NR-Prec: 20.73 -> 19.35 | Improved |
| use classification instead of ranking loss | car model clustering · Rendered cars: car model | NR-Prec: 57.91 -> 56.89 | No clear effect |
Closest papers by meaning
- The Unreasonable Effectiveness of Deep Features as a Perceptual Metric (CVPR 2018 · similarity 0.87)
- Learning With Batch-Wise Optimal Transport Loss for 3D Shape Recognition (CVPR 2019 · similarity 0.87)
- Unsupervised Deep Metric Learning with Transformed Attention Consistency and Contrastive Clustering Loss (ECCV 2020 · similarity 0.87)
- MIC: Mining Interclass Characteristics for Improved Metric Learning (ICCV 2019 · similarity 0.87)
- Sharing Matters for Generalization in Deep Metric Learning (TPAMI 2020 · similarity 0.86)
- Mean Field Theory in Deep Metric Learning (ICLR 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.
