How ultralytics/yolov5 works
YOLOv5 is a PyTorch computer-vision toolkit for object detection, instance segmentation, and image classification. Users provide images or video for inference, or labeled datasets for task-specific training and validation; trained models can then be exported for deployment. The graph also shows the shared model, data, and evaluation layers and the documented Hub and REST interfaces. Relationships are limited to README statements and source-index references; some task and deployment internals were not sampled.
Main entry points: train.py Detection Training · yolo.py YOLO Model · dataloaders.py Image Data · segment/train.py Segmentation Training
Components 25
Vision Tasks
- Detection Inference
detect.py - Classification Inference
classify/predict.py - Segmentation Inference
segment/predict.py - Detection Training
train.py - Classification Training
classify/train.py - Segmentation Training
segment/train.py - Detection Validation
val.py - Classification Validation
classify/val.py - Segmentation Validation
segment/val.py
Shared Runtime
- YOLO Model
models/yolo.py - Model Components
models/common.py - Model Loading
models/experimental.py - Image Data
utils/dataloaders.py - Mask Data
utils/segment/dataloaders.py - Image Augmentation
utils/augmentations.py - Training Losses
utils/loss.py - Evaluation Metrics
utils/metrics.py - Mask Metrics
utils/segment/metrics.py - Runtime Utilities
utils/general.py
Deployment
- Model Export
export.py - PyTorch Hub
hubconf.py - REST API Example
utils/flask_rest_api/restapi.py
Training Tracking
- Training Loggers
utils/loggers/__init__.py - Training Callbacks
utils/callbacks.py
Other components
- User
Connections 33
- User → Detection Inference: runs inference
README.md - User → Classification Inference: runs inference
README.md - User → Segmentation Inference: runs inference
README.md - User → Detection Training: trains model
README.md - User → Model Export: exports model
README.md - Detection Inference → Image Data: loads inputs
- Classification Inference → Image Data: loads inputs
classify/predict.py - Segmentation Inference → Image Data: loads inputs
segment/predict.py - Detection Training → YOLO Model: builds model
train.py - Detection Training → Image Data: loads batches
train.py - Detection Training → Training Losses: computes loss
train.py - Detection Training → Detection Validation: runs validation
train.py - Detection Training → Training Loggers: logs metrics
train.py - Detection Training → Training Callbacks: uses callbacks
train.py - Classification Training → YOLO Model: builds model
classify/train.py - Classification Training → Classification Validation: runs validation
classify/train.py - Segmentation Training → YOLO Model: builds model
segment/train.py - Segmentation Training → Mask Data: loads masks
segment/train.py - Segmentation Training → Mask Metrics: uses metrics
segment/train.py - Segmentation Training → Segmentation Validation: runs validation
segment/train.py - Detection Validation → Image Data: loads batches
val.py - Detection Validation → Evaluation Metrics: computes metrics
val.py - Classification Validation → Image Data: loads batches
classify/val.py - Segmentation Validation → Mask Data: loads masks
segment/val.py - Segmentation Validation → Mask Metrics: computes metrics
segment/val.py - Image Data → Image Augmentation: applies transforms
utils/dataloaders.py - YOLO Model → Model Components: uses layers
models/yolo.py - YOLO Model → Model Loading: uses modules
models/yolo.py - Model Export → YOLO Model: exports architecture
export.py - Classification Inference → YOLO Model: loads model
classify/predict.py - Segmentation Inference → YOLO Model: loads model
segment/predict.py - Detection Training → Runtime Utilities: uses utilities
train.py - Model Export → Runtime Utilities: uses utilities
export.py