How shohruh72/yolov11 works
This repository is a PyTorch YOLOv11 object-detection application. Its principal workflow is training: the command-line entry point builds a dataset and augmentation pipeline, runs the detector and detection loss, then validates and reports metrics. The same entry point also exposes standalone validation and video inference, including detection visualization. The excerpts show the dataset, model, and utility implementations; other documented behavior is represented where supported, without assuming unseen wiring.
Main entry points: main.py Video inference · dataset.py Dataset loading · nn.py YOLO architectures · util.py NMS and metrics
Components 17
Application flows
- CLI entry
main.py - Training loop
main.py - Validation
main.py - Video inference
main.py - Runtime parameters
utils/args.yaml
Data preparation
- Dataset loading
utils/dataset.py - Augmentations
utils/augment.py
Detection model
- YOLO architectures
nets/nn.py - Pretrained weights
weights/v11_n.pt
Training and evaluation
- Detection loss
utils/util.py - EMA updates
utils/util.py - NMS and metrics
utils/util.py - Detection drawing
utils/util.py
Other components
- User
- COCO dataset
- Video source
- Rendered video
Connections 20
- User → CLI entry: selects mode
README.md - CLI entry → Runtime parameters: loads settings
main.py - CLI entry → Training loop: dispatches training
main.py - CLI entry → Validation: dispatches validation
main.py - CLI entry → Video inference: dispatches inference
main.py - Training loop → Dataset loading: creates dataset
main.py - Validation → Dataset loading: uses validation data
main.py - Dataset loading → COCO dataset: reads images and labels
utils/dataset.py - Dataset loading → Augmentations: builds transforms
utils/dataset.py - Training loop → YOLO architectures: trains model
main.py - Training loop → Detection loss: computes loss
main.py - Training loop → EMA updates: updates weights
main.py - Training loop → Validation: validates epochs
main.py - Validation → NMS and metrics: computes metrics
main.py - Video inference → Video source: reads frames
main.py - Video inference → YOLO architectures: runs detector
main.py - Video inference → NMS and metrics: suppresses detections
main.py - Video inference → Detection drawing: draws boxes
main.py - Video inference → Rendered video: writes frames
main.py - Video inference → Pretrained weights: loads pretrained weights