Research map / Label-efficient and robust learning / Recognition and long-tail
Regularizing Deep Networks by Modeling and Predicting Label Structure
Read the paperCheck 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 |
|---|---|---|---|
| Add a learned label-decoder auxiliary regularizer | VGG-16 · PASCAL 2012 val, no pretraining, no augmentation | mIoU: 37.3 -> 42.4 | Helped |
| Add a learned label-decoder auxiliary regularizer | DenseNet-67 · PASCAL 2012 val, no pretraining, no augmentation | mIoU: 40.5 -> 45.2 | Helped |
| Add a learned label-decoder auxiliary regularizer | DenseNet-67 · PASCAL 2012 val, no pretraining, with augmentation | mIoU: 58.8 -> 60.6 | Helped |
| Add a learned label-decoder auxiliary regularizer | VGG-16 · PASCAL 2012 val, ImageNet pretraining, no augmentation | mIoU: 67.1 -> 68.8 | Helped |
| Add a learned label-decoder auxiliary regularizer | DenseNet-121 · PASCAL 2012 val, ImageNet pretraining, with augmentation | mIoU: 71.6 -> 71.9 | Helped |
| Add a learned label-decoder auxiliary regularizer | ResNet-101 · PASCAL 2012 val, ImageNet pretraining, with augmentation | mIoU: 75.4 -> 75.9 | Helped |
| Add a learned label-decoder auxiliary regularizer | DenseNet-67 · PASCAL 2012 val, COCO pretraining, with augmentation | mIoU: 72.3 -> 73.6 | Helped |
| Use encoder-feature prediction instead of the decoder pathway (ablation) | VGG-16 · PASCAL 2012 val, no pretraining, no augmentation | mIoU: 37.3 -> 41.1 | Helped |
| Use encoder-feature prediction instead of the decoder pathway (ablation) | DenseNet-67 · PASCAL 2012 val, no pretraining, no augmentation | mIoU: 40.5 -> 44.0 | Helped |
| Replace the 32-channel bottleneck with a 128-channel bottleneck (ablation) | DenseNet-67 with augmentation, no pretraining · PASCAL 2012 val | mIoU: 59.4 -> 60.6 | Helped |
| Replace the 128-channel bottleneck with a 256-channel bottleneck (ablation) | DenseNet-67 with augmentation, no pretraining · PASCAL 2012 val | mIoU: 60.6 -> 59.8 | Hurt |
| Unfreeze the decoder parameters during CNN training (ablation) | DenseNet-67 + Decoder (128 channel) · PASCAL 2012 val, no pretraining, with augmentation | mIoU: 60.6 -> 60.2 | Hurt |
Closest papers by meaning
- Network as Regularization for Training Deep Neural Networks: Framework, Model and Performance (AAAI 2020 · similarity 0.90)
- Supervised autoencoders: Improving generalization performance with unsupervised regularizers (NeurIPS 2018 · similarity 0.88)
- Simple and Effective Regularization Methods for Training on Noisily Labeled Data with Generalization Guarantee (ICLR 2020 · similarity 0.88)
- Pre-Trained Vision-Language Models as Noisy Partial Annotators (AAAI 2025 · similarity 0.88)
- Learning with Structural Labels for Learning with Noisy Labels (CVPR 2024 · similarity 0.86)
- Automated Label Unification for Multi-Dataset Semantic Segmentation with GNNs (NeurIPS 2024 · similarity 0.86)
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.