Research map / Label-efficient and robust learning / Recognition and long-tail
Rethinking Class-Balanced Methods for Long-Tailed Visual Recognition From a Domain Adaptation Perspective
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Use two-component class-wise and conditional weighting instead of class-balanced cross-entropy | ResNet-32 · CIFAR-LT-10, IF 200 | top-1 error (%): 31.11 -> 29.34 | Helped |
| Use two-component weighting with focal loss | ResNet-32 · CIFAR-LT-10, IF 200 | top-1 error (%): 34.71 -> 25.57 | Helped |
| Use two-component weighting with LDAM | ResNet-32 · CIFAR-LT-10, IF 200 | top-1 error (%): 26.65 -> 22.77 | Helped |
| Use two-component class-wise and conditional weighting instead of class-balanced cross-entropy | ResNet-32 · CIFAR-LT-100, IF 200 | top-1 error (%): 64.30 -> 60.69 | Helped |
| Use two-component weighting instead of class-balanced cross-entropy | ResNet-50 · iNat 2017 | top-1 error (%): 42.59 -> 40.62 | Helped |
| Use two-component weighting instead of class-balanced cross-entropy | ResNet-32 · ImageNet-LT | top-1 error (%): 73.41 -> 70.10 | Helped |
| Use two-component weighting instead of class-balanced cross-entropy | ResNet-152 · Places-LT | top-1 error (%): 71.14 -> 69.20 | Helped |
| Add network pre-training to L2RW (ablation) | L2RW on CIFAR-LT-10 · CIFAR-LT-10, IF 100 | top-1 error (%): 27.77 -> 25.96 | Helped |
| Initialize L2RW sample weights with class-wise weights (ablation) | Pre-trained L2RW on CIFAR-LT-10 · CIFAR-LT-10, IF 100 | top-1 error (%): 25.96 -> 26.26 | Hurt |
| Replace sample weights with class-wise plus conditional weights (ablation) | Pre-trained L2RW on CIFAR-LT-10 · CIFAR-LT-10, IF 100 | top-1 error (%): 25.96 -> 24.54 | Helped |
| Remove clipping and normalization of the weights (ablation) | Two-component L2RW on CIFAR-LT-10 · CIFAR-LT-10, IF 100 | top-1 error (%): 24.54 -> 23.59 | Helped |
| Meta-learn class-wise weights instead of conditional weights (ablation) | Ours on CIFAR-LT-10 · CIFAR-LT-10, IF 100 | top-1 error (%): 23.59 -> 25.42 | Hurt |
Closest papers by meaning
- Decoupling Representation and Classifier for Long-Tailed Recognition (ICLR 2020 · similarity 0.92)
- Balanced Meta-Softmax for Long-Tailed Visual Recognition (NeurIPS 2020 · similarity 0.90)
- Robust Visual Recognition with Class-Imbalanced Open-World Noisy Data (AAAI 2024 · similarity 0.89)
- Decoupled Optimisation for Long-Tailed Visual Recognition (AAAI 2024 · similarity 0.88)
- Feature Fusion from Head to Tail for Long-Tailed Visual Recognition (AAAI 2024 · similarity 0.88)
- Difficulty-aware Balancing Margin Loss for Long-tailed Recognition (AAAI 2025 · similarity 0.88)
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.