Research map / Detection and segmentation / Open-set and few-shot detection
SelvaBox: A high‑resolution dataset for tropical tree crown detection
Hugo Baudchon, Arthur Ouaknine, Martin Weiss, Mélisande Teng, Thomas R. Walla, Antoine Caron-Guay, Christopher Pal, Etienne Laliberté
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 |
|---|---|---|---|
| increase resolution from 10 to 4.5 cm/px (lower GSD) (ablation) | DINO Swin-L · SelvaBox 80x80 m | RF1_75: 44.64 -> 49.38 | Improved |
| use DINO transformer detector instead of Faster R-CNN (ablation) | DINO vs Faster R-CNN ResNet50 · SelvaBox | mAP50:95: 28.74 -> 33.81 (4.5 cm, 80 m) | Improved |
| use Swin L-384 backbone instead of ResNet-50 (ablation) | DINO · SelvaBox | mAP50:95: 33.81 -> 37.79 (4.5 cm, 80 m) | Improved |
| increase input size at fixed 6 cm/px resolution (ablation) | DINO Swin-L · SelvaBox 80x80 m | RF1_75: 47.81 (1333 px) vs 45.88 (1777 px) | No clear effect |
| increase input size at fixed resolution for ResNet-50 models (ablation) | Faster R-CNN / DINO ResNet50 · SelvaBox | mAP50:95: 24.94 -> 27.58 (10 cm, 800 -> 1777 px) | Improved |
| use 40x40 m tiles instead of 80x80 m tiles (ablation) | Faster R-CNN · SelvaBox | RF1_75: 37.52 -> 39.97 | Improved |
| multi-resolution crop and resize augmentation instead of single-resolution training (ablation) | DINO Swin-L · SelvaBox | RF1_75: within standard deviation | No clear effect |
| multi-resolution training vs single-resolution at unseen test resolutions (ablation) | DINO Swin-L · SelvaBox | RF1_75 | Improved |
| add SelvaBox to training data (NQO -> NQOS) (ablation) | DINO Swin-L · SelvaBox test | RF1_75: 30.81 -> 47.63 | Improved |
| add SelvaBox to training data (NQO -> NQOS) (ablation) | DINO Swin-L · BCI50ha zero-shot | RF1_75: 25.32 -> 39.39 | Improved |
| add SelvaBox to training data (NQO -> NQOS) (ablation) | DINO Swin-L · QuebecTrees | RF1_75: 56.53 -> 56.41 | No clear effect |
| train on SelvaBox only instead of NeonTree only (ablation) | DINO Swin-L · QuebecTrees | RF1_75: 14.20 -> 38.34 | Improved |
Closest papers by meaning
- Bringing SAM to new heights: leveraging elevation data for tree crown segmentation from drone imagery (NeurIPS 2025 · similarity 0.88)
- OAM-TCD: A globally diverse dataset of high-resolution tree cover maps (NeurIPS 2024 · similarity 0.87)
- WildBox: A Dataset and Benchmark for Aerial Monocular 3D Detection of African Savanna Wildlife (NeurIPS 2026 · similarity 0.86)
- FoMo: Multi-Modal, Multi-Scale and Multi-Task Remote Sensing Foundation Models for Forest Monitoring (AAAI 2025 · similarity 0.85)
- Open-Canopy: Towards Very High Resolution Forest Monitoring (CVPR 2025 · similarity 0.84)
- Deep Change Monitoring: A Hyperbolic Representative Learning Framework and a Dataset for Long-term Fine-grained Tree Change Detection (CVPR 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.
