Research map / Label-efficient and robust learning / Recognition and long-tail
Simplified Concrete Dropout - Improving the Generation of Attribution Masks for Fine-grained Classification
Dimitri Korsch, Maha Shadaydeh, Joachim Denzler
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 |
|---|---|---|---|
| replace original concrete dropout with the simplified sigmoid formulation | FIDO · CUB-200-2011 SSR mask, ResNet50 | total variation: 39.74 -> 17.54 | Improved |
| replace original concrete dropout with the simplified formulation | FIDO · CUB-200-2011 SDR mask, ResNet50 | total variation: 44.21 -> 22.72 | Improved |
| replace original concrete dropout with the simplified formulation | FIDO · CUB-200-2011 Joint mask, InceptionV3 (iNaturalist2017) | total variation: 28.66 -> 14.06 | Improved |
| use simplified concrete dropout with mini-batch size 8 instead of 32 | FIDO · CUB-200-2011 mask estimation, 100 steps | computation time per image: 40 -> 11 seconds | Improved |
| use simplified concrete dropout (smaller mini-batches) | FIDO · Red-winged and Yellow-headed Blackbird masks | IoU with ground-truth mask | Improved |
| add FIDO joint-mask crop as test-time augmentation | ResNet50 · CUB-200-2011 | accuracy (%): 82.78 -> 84.17 | Improved |
| add simplified-CD FIDO crop instead of original FIDO crop | ResNet50 · CUB-200-2011 | accuracy (%): 84.17 -> 84.67 | Improved |
| add simplified-CD FIDO crop instead of original FIDO crop | InceptionV3 (iNaturalist2017) · CUB-200-2011 | accuracy (%): 90.47 -> 90.51 | No clear effect |
| add random crop test-time augmentation | InceptionV3 (iNaturalist2017) · CUB-200-2011 | accuracy (%): 90.32 -> 89.99 | Hurt |
| add center crop test-time augmentation | ResNet50 · CUB-200-2011 | accuracy (%): 82.78 -> 83.83 | Improved |
| add gradient-based crop test-time augmentation | InceptionV3 · CUB-200-2011 | accuracy (%): 79.86 -> 80.76 | Improved |
| add BirdYolo detector crop test-time augmentation | InceptionV3 · CUB-200-2011 | accuracy (%): 79.86 -> 81.12 | Improved |
Closest papers by meaning
- Fine-Grained Image Analysis With Deep Learning: A Survey (TPAMI 2021 · similarity 0.85)
- Twofold Debiasing Enhances Fine-Grained Learning with Coarse Labels (AAAI 2025 · similarity 0.84)
- Fine-grained Classes and How to Find Them (ICML 2024 · similarity 0.84)
- Cross-Layer and Cross-Sample Feature Optimization Network for Few-Shot Fine-Grained Image Classification (AAAI 2024 · similarity 0.83)
- Multi-scale Activation, Refinement, and Aggregation: Exploring Diverse Cues for Fine-Grained Bird Recognition (AAAI 2025 · similarity 0.83)
- Advancing Fine-Grained Classification by Structure and Subject Preserving Augmentation (NeurIPS 2024 · similarity 0.83)
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.
