Research map / Detection and segmentation / CNN features
RecursiveMix: Mixed Learning with History
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 |
|---|---|---|---|
| Use RecursiveMix instead of CutMix | DenseNet-161 · CIFAR-100 | Top-1 error: 16.64 -> 15.54 | Helped |
| Use RecursiveMix instead of CutMix + MoEx | ResNet-50 · ImageNet validation | Top-1 error: 20.90 -> 20.80 | Helped |
| Pretrain with RecursiveMix instead of CutMix | ResNet-50 with ATSS · COCO | AP: 40.1 -> 41.5 | Helped |
| Pretrain with RecursiveMix instead of CutMix | ResNet-50 with UperNet · ADE20K | mIoU: 41.24 -> 42.30 | Helped |
| Add resize strategy to CutMix (ablation) | PyramidNet-164 · CIFAR-100 | Top-1 error: 16.67 -> 15.36 | Helped |
| Add historical mixing to the resize strategy (ablation) | PyramidNet-164 · CIFAR-100 | Top-1 error: 15.36 -> 14.81 | Helped |
| Add consistency loss to resize and historical mixing (ablation) | PyramidNet-164 · CIFAR-100 | Top-1 error: 14.81 -> 14.65 | Helped |
| Use separate historical and current classifier weights (ablation) | RM · CIFAR-10 | Top-1 error: 2.61 -> 2.35 | Helped |
| Use separate historical and current classifier weights (ablation) | RM · ImageNet validation | Top-1 error: 20.92 -> 20.80 | Helped |
| Use nearest instead of bilinear interpolation (ablation) | RM · CIFAR-10 | Top-1 error: 2.58 -> 2.35 | Helped |
| Use nearest instead of bilinear interpolation (ablation) | RM · ImageNet validation | Top-1 error: 20.86 -> 20.80 | Helped |
| Vary α across the tested values (ablation) | RM · CIFAR-10 and CIFAR-100 | Top-1 error | Mixed |
Closest papers by meaning
- Catch-Up Mix: Catch-Up Class for Struggling Filters in CNN (AAAI 2024 · similarity 0.86)
- Supervision Interpolation via LossMix: Generalizing Mixup for Object Detection and Beyond (AAAI 2024 · similarity 0.86)
- Adversarial AutoMixup (ICLR 2024 · similarity 0.86)
- DiffuseMix: Label-Preserving Data Augmentation with Diffusion Models (CVPR 2024 · similarity 0.85)
- ReCon: Region-Controllable Data Augmentation with Rectification and Alignment for Object Detection (NeurIPS 2025 · similarity 0.85)
- LoMix: Learnable Weighted Multi-Scale Logits Mixing for Medical Image Segmentation (NeurIPS 2025 · 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.