Research map / Point clouds / Segmentation and detection
SegFly: A 2D-3D-2D Paradigm for Aerial RGB-Thermal Semantic Segmentation at Scale
Markus Gross, Sai Bharadhwaj Matha, Rui Song, Viswanathan Muthuveerappan, Conrad Christoph, Julius Huber, Daniel Cremers
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 |
|---|---|---|---|
| train on SegFly-RGB pseudo-labels instead of manual GT only | Firefly · SegFly-RGB test, eval on manual GT | FWmIoU (%): 45.81 -> 54.90 | Improved |
| train on SegFly-RGB pseudo-labels instead of manual GT only | UPerNet · SegFly-RGB test, eval on manual GT | FWmIoU (%): 52.46 -> 54.36 | Improved |
| train on SegFly-RGB pseudo-labels instead of manual GT only | SegFormer · SegFly-RGB test, eval on manual GT | FWmIoU (%): 48.50 -> 51.36 | Improved |
| add RGB pretraining before RGB-T finetuning (ablation) | Firefly · SegFly-RGB-T test | FWmIoU (%): 35.20 -> 37.38 | Improved |
| add RGB-T Rein adaptation stage (ablation) | Firefly · SegFly-RGB-T test | FWmIoU (%): 37.38 -> 43.83 | Improved |
| add RGB-T Rein adaptation stage (ablation) | UPerNet · SegFly-RGB-T test | Acc. / FWmIoU (%): 33.48 -> 34.74 Acc; 32.21 -> 31.98 FWmIoU | Mixed |
| add RGB features by concatenation (RGB-T fusion) (ablation) | Firefly · SegFly-RGB-T test | Acc./FWmIoU (%): +1.27/+0.43 | Improved |
| add occlusion filtering (ablation) | point-based rendering · test scene 8 | accuracy (%): 89.78 -> 90.08 | Improved |
| add splatting-based densification (ablation) | point-based rendering · test scene 8 | accuracy (%): 89.54 -> 90.08 | Improved |
| add depth-guided label distribution (ablation) | point-based rendering · test scene 8 | accuracy (%): 83.95 -> 90.08 | Improved |
| add thermal lens distortion transfer (ablation) | RGB-T registration · registration accuracy/FWmIoU | accuracy (%): 84.08 -> 87.05 | Improved |
| increase manual annotation from 1.1% to 3.1% (ablation) | 2D-3D-2D pseudo-labels · test scene 9, RGB | accuracy (%): 73.96 -> 87.53 | Improved |
Closest papers by meaning
- Unleashing Multispectral Video's Potential in Semantic Segmentation: A Semi-supervised Viewpoint and New UAV-View Benchmark (NeurIPS 2024 · similarity 0.88)
- Aerial Lifting: Neural Urban Semantic and Building Instance Lifting from Aerial Imagery (CVPR 2024 · similarity 0.87)
- High Temporal Consistency through Semantic Similarity Propagation in Semi-Supervised Video Semantic Segmentation for Autonomous Flight (CVPR 2025 · similarity 0.86)
- ThermalGen: Style-Disentangled Flow-Based Generative Models for RGB-to-Thermal Image Translation (NeurIPS 2025 · similarity 0.86)
- PixDLM: A Dual-Path Multimodal Language Model for UAV Reasoning Segmentation (CVPR 2026 · similarity 0.86)
- SemAlign3D: Semantic Correspondence between RGB-Images through Aligning 3D Object-Class Representations (CVPR 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.
