Research map / Detection and segmentation / CNN features
DejaVu: Conditional Regenerative Learning To Enhance Dense Prediction
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 conditional regeneration loss | HRNet18 segmentation · Cityscapes val | mIoU: 77.6 -> 78.8 | Helped |
| Add conditional regeneration loss | Mask2Former with Swin-T · COCO val panoptic | PQ: 53.2 -> 54.3 | Helped |
| Add conditional regeneration loss | DenseCLIP with ViT-B · ADE20K val | mIoU: 49.8 -> 50.3 | Helped |
| Add conditional regeneration loss | Monodepth2-R50 · KITTI eigen split | Sq Rel: 0.903 -> 0.769 | Helped |
| Add conditional regeneration loss | XTC multi-task model · NYUD-v2 fully supervised | Seg. mIoU: 41.00 -> 42.69 | Helped |
| Use spatial rather than no redaction (ablation) | Single-task dense prediction · NYUD-v2 | Seg. mIoU: 37.25 -> 38.38 | Helped |
| Use spectral rather than no redaction (ablation) | Single-task dense prediction · NYUD-v2 | Depth aErr: 59.70 -> 56.76 | Helped |
| Use spatial rather than spectral redaction (ablation) | Single-task dense prediction · NYUD-v2 | Seg. mIoU: 38.21 -> 38.38 | Helped |
| Use spectral rather than spatial redaction (ablation) | Single-task dense prediction · NYUD-v2 | Depth aErr: 58.34 -> 56.76 | Helped |
| Add CLIP text supervision to DejaVu (ablation) | HRNet18 segmentation · Cityscapes val | mIoU: 78.8 -> 79.2 | Helped |
| Add CLIP text supervision to DejaVu (ablation) | HRNet18-OCR segmentation · Cityscapes val | mIoU: 82.0 -> 82.3 | Helped |
| Remove the regeneration objective | Dense prediction training · multiple dense prediction benchmarks | dense prediction performance | Hurt |
Closest papers by meaning
- Lotus: Diffusion-based Visual Foundation Model for High-quality Dense Prediction (ICLR 2025 · similarity 0.86)
- Decompositional Neural Scene Reconstruction with Generative Diffusion Prior (CVPR 2025 · similarity 0.84)
- Exploiting Diffusion Prior for Real-World Image Dehazing with Unpaired Training (AAAI 2025 · similarity 0.84)
- Win-Win: Training High-Resolution Vision Transformers from Two Windows (ICLR 2024 · similarity 0.83)
- Adaptive Depth Networks with Skippable Sub-Paths (NeurIPS 2024 · similarity 0.83)
- DRScaffold: Boosting Dense-Scene Reasoning in Lightweight Vision Language Models (NeurIPS 2026 · 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.