Research map / Label-efficient and robust learning / Domain adaptation and generalization
Progressive Random Convolutions for Single Domain Generalization
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 |
|---|---|---|---|
| Replace single-layer RandConv with progressive small-kernel convolutions and a random convolution block | LeNet · Digits (MNIST source, target-domain average) | accuracy (%): 74.84 -> 81.35 | Helped |
| Replace single-layer RandConv with progressive small-kernel convolutions and a random convolution block | ResNet18 · PACS single DG (target-domain average) | accuracy (%): 67.50 -> 68.88 | Helped |
| Replace single-layer RandConv with progressive small-kernel convolutions and a random convolution block | ResNet18 · PACS multi DG (target-domain average) | accuracy (%): 82.43 -> 84.29 | Helped |
| Replace single-layer RandConv with progressive small-kernel convolutions and a random convolution block | ResNet50 · PACS multi DG (target-domain average) | accuracy (%): 86.63 -> 88.27 | Helped |
| Use Pro-RandConv instead of RandConv | DeepLabV3+ with ResNet50 · GTAV to Cityscapes, BDD-100K, SYNTHIA, Mapillary (average) | mIoU (%): 30.80 -> 36.19 | Helped |
| Use the same random weights across progressive convolution layers rather than different weights (ablation) | Progressive random convolutions · Digits SDG | accuracy (%): 74.72 -> 78.26 | Helped |
| Use the same random weights across progressive convolution layers rather than different weights (ablation) | Progressive random convolutions · PACS single DG | accuracy (%): 66.47 -> 68.39 | Helped |
| Add Gaussian smoothing to the shared-weight progressive approach (ablation) | Pro-RandConv · Digits SDG | accuracy (%): 78.26 -> 80.03 | Helped |
| Add contrast diversification to the progressive approach (ablation) | Pro-RandConv · PACS single DG | accuracy (%): 67.89 -> 68.73 | Helped |
| Add deformable offsets to the progressive approach (ablation) | Pro-RandConv · PACS single DG | accuracy (%): 67.89 -> 68.25 | Helped |
| Add deformable offsets to the full segmentation augmentation (ablation) | Pro-RandConv segmentation · GTAV to Cityscapes, BDD-100K, SYNTHIA, Mapillary (average) | mIoU (%): 36.64 -> 36.19 | Hurt |
| Train using only augmented images rather than original and augmented images (ablation) | Pro-RandConv · PACS single DG (out-of-domain) | accuracy (%): 68.88 -> 68.10 | No clear effect |
Closest papers by meaning
- START: A Generalized State Space Model with Saliency-Driven Token-Aware Transformation (NeurIPS 2024 · similarity 0.86)
- CLIPCEIL: Domain Generalization through CLIP via Channel rEfinement and Image-text aLignment (NeurIPS 2024 · similarity 0.85)
- Beyond Random Augmentations: Pretraining with Hard Views (ICLR 2025 · similarity 0.85)
- Unbiased Faster R-CNN for Single-source Domain Generalized Object Detection (CVPR 2024 · similarity 0.85)
- DGFamba: Learning Flow Factorized State Space for Visual Domain Generalization (AAAI 2025 · similarity 0.84)
- PracticalDG: Perturbation Distillation on Vision-Language Models for Hybrid Domain Generalization (CVPR 2024 · similarity 0.84)
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.