CNNs outperform Transformers on tree canopy segmentation with limited data

Researchers investigated the effectiveness of five different deep learning architectures, including YOLOv11, Mask R-CNN, DeepLabv3, Swin-UNet, and DINOv2, for tree canopy segmentation using a very limited dataset of only 150 images. Their findings indicate that pretrained convolutional neural network models, specifically YOLOv11 and Mask R-CNN, demonstrated superior generalization compared to transformer-based models in this low-data scenario. The study suggests that transformer architectures struggle with extreme data scarcity without extensive pretraining or augmentation, and highlights the continued reliability of lightweight CNNs for tasks with limited imagery. AI

IMPACT Demonstrates that CNNs remain competitive for specialized tasks with scarce data, potentially guiding model selection for similar applications.

RANK_REASON Academic paper evaluating model performance on a specific task with limited data. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CV →

• arXiv
• DeepLabv3
• DINOv2
• Mask R-CNN
• Solafune Tree Canopy Detection competition
• Swin-UNet
• YOLOv11

• paper
• other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →