Research map / Image generation / Autoregressive and latent generation
Task-Oriented Data Synthesis and Control-Rectify Sampling for Remote Sensing Semantic Segmentation
Yunkai Yang, Yudong Zhang, Kunquan Zhang, Jinxiao Zhang, Xinying Chen, Haohuan Fu, Runmin Dong
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 |
|---|---|---|---|
| add control-rectify flow matching (CRFM) sampling | SD v3.5 Tri-attention · FUSU-4k | mIoU: 48.57 -> 49.41 | Improved |
| add CRFM sampling | SD v3.5 Tri-attention · LoveDA-5k (rural-to-urban) | mIoU: 65.45 -> 66.12 | Improved |
| train on real plus TODSynth data vs real only | segmentation baseline · FUSU-4k | mIoU: 45.27 -> 49.41 | Improved |
| train on real plus TODSynth data vs real only | segmentation baseline · LoveDA-5k (rural-to-urban) | mIoU: 64.04 -> 66.12 | Improved |
| use Tri-attention instead of mask-adapter (ablation) | SD v3.5 · FUSU-4k | mIoU: 47.41 -> 48.57 | Improved |
| use Tri-attention instead of siamese MM-attention (ablation) | SD v3.5 · FUSU-4k | mIoU: 48.46 -> 48.57 | Improved |
| use FreeMask pixel-level filter instead of CLIP score filter (ablation) | SD v3.5 · FUSU-4k | mIoU: 46.69 -> 48.57 | Improved |
| increase CRFM steps from 0 to 2 (ablation) | SD v3.5 (23 steps) · FUSU-4k | mIoU: 48.57 -> 48.80 | No clear effect |
| increase CRFM steps from 4 to 6 (ablation) | SD v3.5 (23 steps) · FUSU-4k | FID: 38.65 -> 66.95 | Hurt |
| increase CRFM steps from 4 to 6 (ablation) | SD v3.5 (23 steps) · FUSU-4k | mIoU: 49.41 -> 48.74 | Hurt |
| apply CRFM with 6 steps at 18 sampling steps (ablation) | SD v3.5 (18 steps) · FUSU-4k | mIoU: 47.81 -> 49.57 | Mixed |
| optimise latent z_t directly with segmentation gradient instead of velocity field (ablation) | CRFM · FUSU | synthetic data effectiveness | Hurt |
Closest papers by meaning
- SatSynth: Augmenting Image-Mask Pairs through Diffusion Models for Aerial Semantic Segmentation (CVPR 2024 · similarity 0.88)
- MMM-RS: A Multi-modal, Multi-GSD, Multi-scene Remote Sensing Dataset and Benchmark for Text-to-Image Generation (NeurIPS 2024 · similarity 0.86)
- ROS-SAM: High-Quality Interactive Segmentation for Remote Sensing Moving Object (CVPR 2025 · similarity 0.86)
- SemStereo: Semantic-Constrained Stereo Matching Network for Remote Sensing (AAAI 2025 · similarity 0.86)
- SemFlow: Binding Semantic Segmentation and Image Synthesis via Rectified Flow (NeurIPS 2024 · similarity 0.86)
- Towards Open-Vocabulary Remote Sensing Image Semantic Segmentation (AAAI 2025 · similarity 0.85)
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.
