How weecology/deepforest works
Project type and purpose
• DeepForest is an open-source Python package (library/CLI tool) for object-detection in aerial imagery, built on PyTorch/torchvision. Its core functionality is training and running tree-crown and bird detection models on RGB images, plus evaluation and visualization tools.
File structure
• Top-level:
• src/deepforest/… – core library code (data handling, models, training, prediction, evaluation, visualization, CLI).
• src/deepforest/models/… – model definitions (FasterRCNN, RetinaNet).
• src/deepforest/datasets/… – dataset builders for training and prediction.
• scripts/cli.py & main.py – command-line interface entrypoints.
• conf/config.yaml – default configuration.
• download.py – fetching pretrained weights.
• tests/… – unit tests for each component.
• docs/, README.md, CI workflows – documentation & deployment assets.
README
• Confirms: DeepForest uses torchvision object-detection modules.
• Key features: model loading, custom training, prediction, evaluation, visualization.
• Dependencies: Python ≥3.8, PyTorch, torchvision.
Detail
• Config file feeding into both data and model modules.
• Cache/download service for pretrained weights.