Raidenop/solar-panels-rtdetr

Solar Panel Detection with RT-DETR

Fine-tuned RT-DETR (Real-Time Detection Transformer) for detecting solar panels and their conditions in aerial/satellite imagery.

Model Details

• Base Model: PekingU/rtdetr_r50vd_coco_o365 (pretrained on COCO + Objects365)
• Architecture: RT-DETR with ResNet-50-vd backbone
• Task: Object Detection (bounding box prediction)
• Image Size: 640×640
• Classes (6): solar-panels, Cell, Cell-Multi, No-Anomaly, Shadowing, Unclassified

Training Details

• Dataset: Francesco/solar-panels-taxvb (112 train / 19 val / 30 test images)
• Epochs: 100
• Batch Size: 8
• Learning Rate: 5e-5 (cosine schedule with 5% warmup)
• Optimizer: AdamW (weight_decay=1e-4)
• Gradient Clipping: max_grad_norm=0.01
• Precision: FP16
• Augmentations: Perspective, HorizontalFlip, VerticalFlip, RandomBrightnessContrast, HueSaturationValue, RandomRotate90

Usage

from transformers import AutoImageProcessor, AutoModelForObjectDetection
from PIL import Image
import torch

# Load model and processor
model = AutoModelForObjectDetection.from_pretrained("Raidenop/solar-panels-rtdetr")
processor = AutoImageProcessor.from_pretrained("Raidenop/solar-panels-rtdetr")

# Load image
image = Image.open("your_image.jpg")

# Inference
with torch.no_grad():
    inputs = processor(images=[image], return_tensors="pt")
    outputs = model(**inputs)
    target_sizes = torch.tensor([[image.size[1], image.size[0]]])
    results = processor.post_process_object_detection(
        outputs, threshold=0.3, target_sizes=target_sizes
    )[0]

# Print results
for score, label, box in zip(results["scores"], results["labels"], results["boxes"]):
    box = [round(i, 2) for i in box.tolist()]
    print(f"Detected {model.config.id2label[label.item()]} with confidence {round(score.item(), 3)} at {box}")

Training Script

The full training script is available in this repository: train.py

To run training:

pip install transformers datasets accelerate torch torchvision trackio albumentations>=1.4.5 torchmetrics pycocotools timm
python train.py

Requirements: NVIDIA GPU with 16GB+ VRAM (T4, A10G, or better)

References

• RT-DETR Paper — "DETRs Beat YOLOs on Real-time Object Detection" (CVPR 2023)
• HF Object Detection Tutorial
• Dataset: solar-panels-taxvb (Roboflow 100 benchmark)

Generated by ML Intern

This model repository was generated by ML Intern, an agent for machine learning research and development on the Hugging Face Hub.

• Try ML Intern: https://smolagents-ml-intern.hf.space
• Source code: https://github.com/huggingface/ml-intern