NEON Tree Genus Classification - RESNET18

A resnet18 model trained for tree genus classification on the NEON Tree Crown Dataset. This model is designed for integration with DeepForest as a CropModel.

Model Details

Architecture: resnet18
Task: Tree genus classification
Classes: 60 genus classes
Input size: 224x224 RGB images
Normalization: ImageNet (mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
Dataset: NEON Tree Crown Dataset (~48,000 tree crowns from 30 NEON sites)

Usage with DeepForest

from deepforest import CropModel

# Load model
model = CropModel.load_model("ritesh313/neon-tree-resnet18-genus")

# Use with DeepForest predictions
# (after running detection with main DeepForest model)
results = model.predict(image_crops)

Direct PyTorch Usage

import torch
from safetensors.torch import load_file
from torchvision import transforms

# Load model weights
state_dict = load_file("model.safetensors")

# Load config for label mapping
import json
with open("config.json") as f:
    config = json.load(f)

# Create your model architecture and load weights
# model.load_state_dict(state_dict)

# Preprocessing
preprocess = transforms.Compose([
    transforms.Resize((224, 224)),
    transforms.ToTensor(),
    transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
])

Training Details

Framework: PyTorch Lightning
Optimizer: AdamW
Learning Rate: 1e-3
Scheduler: ReduceLROnPlateau
Data Split: 70/15/15 (train/val/test)
Seed: 42

Dataset

The model was trained on the NEON Tree Crown Dataset, which includes:

47,971 individual tree crowns
167 species / 60 genera
30 NEON sites across North America
Multi-modal data: RGB, Hyperspectral, LiDAR (this model uses RGB only)

Citation

If you use this model, please cite:

@software{neontreeclassification,
  author = {Chowdhry, Ritesh},
  title = {NeonTreeClassification: Multi-modal Tree Species Classification},
  url = {https://github.com/Ritesh313/NeonTreeClassification},
  year = {2026}
}

License

MIT License

ritesh313/neon-tree-resnet18-genus