How lingtengqiu/yolo_nano works
PyTorch-YOLO_Nano is an object-detection implementation supporting model training, evaluation/testing, and inference over image or video datasets. The main workflow configures a dataset and model, trains with augmentation and learning-rate scheduling, and saves checkpoints; inference loads weights, runs detection, and writes annotated video. The graph includes the documented augmentation and training features. Some connections are based on SOURCE INDEX references rather than complete excerpts, so uncertain call paths are omitted.
Main entry points: train.py Train model · datasets.py Dataset loading · inference.py Run inference · darknet.py Darknet network
Components 18
Product workflows
- Train model
train.py - Test model
test.py - Run inference
inference.py - Detect objects
detect.py - Annotated video
result
Data preparation
- Dataset loading
utils/datasets.py - Dataset catalog
data/default_path.py - Image augmentation
utils/augmentations.py - Memcached reader
utils/mc_reader.py
Detection model
- Darknet network
models/darknet.py - YOLO Nano network
models/yolo_nano_helper.py - Config parsing
utils/parse_config.py - Detection utilities
utils/utils.py
Training support
- Learning-rate policy
utils/optim.py - Training logger
utils/logger.py - TensorBoard visualizer
utils/board.py - Saved checkpoints
Other components
- User
Connections 27
- User → Train model: starts training
README.md - User → Test model: runs tests
README.md - User → Run inference: starts inference
- User → Detect objects: starts detection
- Train model → Config parsing: parses config
train.py - Train model → Dataset loading: loads batches
train.py - Train model → Darknet network: trains network
train.py - Train model → YOLO Nano network: trains network
train.py - Train model → Learning-rate policy: sets learning rate
train.py - Train model → Training logger: logs metrics
train.py - Train model → Saved checkpoints: saves weights
train.py - Train model → Test model: evaluates model
train.py - Dataset loading → Dataset catalog: gets dataset paths
utils/datasets.py - Dataset loading → Image augmentation: augments images
utils/datasets.py - Dataset loading → Memcached reader: reads cached data
utils/datasets.py - Darknet network → Config parsing: parses model config
models/darknet.py - Darknet network → Detection utilities: builds targets
models/darknet.py - Test model → Dataset loading: loads evaluation data
- Test model → Darknet network: evaluates network
- Run inference → Dataset loading: loads inputs
inference.py - Run inference → Darknet network: runs network
inference.py - Run inference → YOLO Nano network: runs network
inference.py - Run inference → Detection utilities: processes detections
inference.py - Run inference → Annotated video: writes video
inference.py - Detect objects → Dataset loading: loads images
detect.py - Detect objects → Detection utilities: uses detection helpers
detect.py - TensorBoard visualizer → Training logger: records metrics