Research map / Label-efficient and robust learning / Recognition and long-tail
Hierarchy of Alternating Specialists for Scene Recognition
Hyo Jin Kim, Jan-Michael Frahm
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 class-based coarse grouping (HD-CNN*) with confusing-cluster specialists | AlexNet* hierarchy · SUN-190 | Top-1 accuracy (%): 63.65 -> 66.41 | Improved |
| add hierarchy levels (Lev 1 to Lev 3) with alternating architecture (ablation) | AlexNet* specialists · SUN-190 | Top-1 accuracy (%): 66.13 -> 66.41 | Improved |
| add hierarchy levels (Lev 1 to Lev 3) with alternating architecture (ablation) | AlexNet* specialists · Places-205 | Top-1 accuracy (%): 50.21 -> 51.54 | Improved |
| add alternating architecture instead of ordered pooling only (ablation) | hierarchy of specialists · SUN-190 Lev 3 | Top-1 accuracy (%): 64.43 -> 66.41 | Improved |
| add alternating architecture instead of orderless pooling only (ablation) | hierarchy of specialists · SUN-190 Lev 3 | Top-1 accuracy (%): 63.08 -> 66.41 | Improved |
| add alternating architecture instead of ordered pooling only (ablation) | hierarchy of specialists · Places-205 Lev 3 | Top-1 accuracy (%): 50.28 -> 51.54 | Improved |
| add alternating architecture instead of orderless pooling only (ablation) | hierarchy of specialists · CIFAR-100 Lev 3 | Top-1 accuracy (%): 66.62 -> 67.70 | Improved |
| use alternating specialists instead of late fusion of both representations | AlexNet* generalist · SUN-190 | Top-1 accuracy (%): 64.45 -> 66.41 | Improved |
| add specialist hierarchy to base NIN-C100 | NIN-C100 · CIFAR-100 | Top-1 accuracy (%): 64.73 -> 67.70 | Improved |
| add specialist hierarchy to base model | AlexNet* · Places-205 | Top-1 accuracy (%): 48.67 -> 51.54 | Improved |
| increase HD-CNN* number of clusters from 2 to 8 (ablation) | HD-CNN* · SUN-190 | Top-1 accuracy (%): 63.11 -> 63.65 top-1; top-5 88.81 -> 84.08 | Mixed |
| increase ensemble size from 2 to 8 models (ablation) | simple ensemble · SUN-190 | Top-1 accuracy (%): 64.19 -> 64.99 | Improved |
Closest papers by meaning
- Network of Experts for Large-Scale Image Categorization (ECCV 2016 · similarity 0.87)
- Learning from Small Sample Sets by Combining Unsupervised Meta-Training with CNNs (NeurIPS 2016 · similarity 0.85)
- Rethinking Zero-Shot Learning: A Conditional Visual Classification Perspective (ICCV 2019 · similarity 0.85)
- Large Scale Visual Recognition through Adaptation using Joint Representation and Multiple Instance Learning (JMLR 2016 · similarity 0.85)
- Hierarchical Novelty Detection for Visual Object Recognition (CVPR 2018 · similarity 0.85)
- CAR: Class-Aware Regularizations for Semantic Segmentation (ECCV 2022 · 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.
