How zsniko/ednet works
EDNet is an edge-oriented small-object detector for UAV imagery, built on an Ultralytics-style YOLO framework. The principal workflow is to construct a model, prepare and augment labeled data, train it, validate detection quality, run predictions and inspect results, then export for deployment. The README also describes dataset exploration and mobile/CoreML use. Paths and links reflect the supplied tree, excerpts, and documented API; unsampled implementation details are represented only where the tree or README supports them.
Main entry points: __init__.py EDNet API · trainer.py Training engine · build.py Dataset builder · tasks.py Task model
Components 23
Model API
- EDNet API
ednet/__init__.py - YOLO model
ednet/models/yolo/model.py
Data preparation
- Dataset builder
ednet/data/build.py - Dataset
ednet/data/dataset.py - Data loaders
ednet/data/loaders.py - Augmentation
ednet/data/augment.py
Model runtime
- Task model
ednet/nn/tasks.py - Network blocks
ednet/nn/modules/block.py - Training engine
ednet/engine/trainer.py - Detection training
ednet/models/yolo/detect/train.py - Training losses
ednet/utils/loss.py
Evaluation and deployment
- Validation engine
ednet/engine/validator.py - Detection validation
ednet/models/yolo/detect/val.py - Evaluation metrics
ednet/utils/metrics.py - Prediction engine
ednet/engine/predictor.py - Detection prediction
ednet/models/yolo/detect/predict.py - Prediction results
ednet/engine/results.py - Model exporter
ednet/engine/exporter.py
Data exploration
- Dataset explorer
ednet/data/explorer/explorer.py
Other components
- Engineer
- LanceDB
- Comet ML
- CoreML runtime
Connections 26
- Engineer → EDNet API: uses
- EDNet API → YOLO model: constructs
- YOLO model → Task model: builds task
- Task model → Network blocks: assembles
- EDNet API → Training engine: trains
- Training engine → Dataset builder: builds data
- Dataset builder → Dataset: creates
- Dataset builder → Data loaders: creates loaders
- Dataset → Augmentation: transforms with
- Training engine → Detection training: dispatches
- Detection training → Task model: trains
- Detection training → Training losses: computes loss
- EDNet API → Validation engine: validates
- Validation engine → Dataset builder: builds data
- Validation engine → Detection validation: dispatches
- Detection validation → Evaluation metrics: updates
- EDNet API → Prediction engine: predicts
- Prediction engine → Detection prediction: dispatches
- Detection prediction → Task model: runs inference
- Detection prediction → Prediction results: returns
- EDNet API → Model exporter: exports
- Model exporter → CoreML runtime: exports for
- Engineer → Dataset explorer: explores data
- Dataset explorer → Dataset: reads images
- Dataset explorer → LanceDB: writes embeddings
- Training engine → Comet ML: logs experiments