Research map / Image generation / Super-resolution
Mining self-similarity: Label super-resolution with epitomic representations
Nikolay Malkin, Anthony Ortiz, Nebojsa Jojic
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 U-Net with epitome, both trained on South HR labels | South-trained segmentation · Chesapeake land cover (North region) | overall accuracy (%): 59.4 -> 79.5 | Improved |
| increase self-epitomic tile size from 256^2 to 2048^2 (ablation) | self-epitomic LSR · Chesapeake land cover (North region) | overall accuracy (%): 85.9 -> 88.0 | Improved |
| increase self-epitomic tile size from 256^2 to 2048^2 (ablation) | self-epitomic LSR · Chesapeake land cover (North region) | mean IoU (%): 63.3 -> 67.8 | Improved |
| use whole-region (all-tile) epitome instead of 2048^2 tile (ablation) | epitomic LSR · Chesapeake land cover (North region) | overall accuracy (%): 88.0 -> 83.9 | Hurt |
| use 512^2 self-epitomic LSR instead of neural LSR trained with HR (S) and LR (N) | LSR · Chesapeake land cover (North region) | overall accuracy (%): 86.9 -> 87.0 | Improved |
| use epitomic LSR instead of U-Net neural LSR with LR (North) only | LSR · Chesapeake land cover (North region) | mean IoU (%): 41.3 -> 58.5 | Improved |
| add self-diversification to epitomic LSR (ablation) | epitomic LSR · TIL segmentation (1786 labeled cell images) | AUC: 0.794 -> 0.801 | Improved |
| use diversified epitomic LSR instead of CNN with pretraining | TIL detection · TIL segmentation (1786 labeled cell images) | AUC: 0.786 -> 0.801 | Improved |
| ensemble color-only model with epitome model on shrubland and barren (ablation) | GRSS LSR · GRSS 2020 Data Fusion validation | average accuracy (%): 65.5 -> 68 | Improved |
| train small FCN on ensemble predictions (ablation) | GRSS LSR · GRSS 2020 Data Fusion validation | average accuracy (%): 68 -> 70.7 | Improved |
| use self-epitomic LSR instead of U-Net inference | land cover · Land cover inference time | ms per thousand pixels: 3 -> 94 | Hurt |
Closest papers by meaning
- Label super-resolution networks (ICLR 2019 · similarity 0.88)
- Label-Efficient Few-Shot Semantic Segmentation with Unsupervised Meta-Training (AAAI 2024 · similarity 0.83)
- Learned Representation-Guided Diffusion Models for Large-Image Generation (CVPR 2024 · similarity 0.83)
- SeD: Semantic-Aware Discriminator for Image Super-Resolution (CVPR 2024 · similarity 0.83)
- CuVLER: Enhanced Unsupervised Object Discoveries through Exhaustive Self-Supervised Transformers (CVPR 2024 · similarity 0.83)
- Exploring Simple Open-Vocabulary Semantic Segmentation (CVPR 2025 · similarity 0.83)
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.
