COR Brief
Image & Video

YOLOv3

YOLOv3 is an object detection algorithm developed by Joseph Redmon that performs real-time detection by predicting bounding boxes and class probabilities directly from full images in a single evaluation pass. It introduces multiscale predictions using three different detection kernel sizes, achieving 28.2 mAP on the COCO dataset while running at 22 milliseconds per frame on 320×320 input. This performance matches the accuracy of SSD but operates approximately three times faster. Ultralytics provides a PyTorch implementation of YOLOv3 that supports forward compatibility with YOLOv5 models and methods, including exporting models to ONNX, CoreML, and TFLite formats. The model can be trained on datasets such as COCO using Python or command-line interfaces.

Updated Jan 28, 2026open-source

YOLOv3 is a real-time object detection model that balances accuracy and speed using multiscale predictions and a single neural network evaluation.

Pricing
open-source
Category
Image & Video
Company
Interactive PresentationOpen Fullscreen ↗
01
Performs detection using three different kernel sizes to improve accuracy across object scales.
02
Supports inference via PyTorch Hub and export to ONNX, CoreML, and TFLite formats for deployment on various platforms.
03
Enables training on datasets like COCO using Ultralytics API or CLI tools.
04
Maintains compatibility with updates and methods from YOLOv5 for ongoing maintenance and improvements.

Real-Time Object Detection

Detect objects in video streams or images with low latency for applications such as surveillance or robotics.

Edge Device Deployment

Export models to lightweight formats like TFLite or CoreML for deployment on mobile or embedded devices.

Custom Dataset Training

Train the model on specific datasets to tailor detection capabilities to specialized domains.

1
Clone Repository
Run git clone https://github.com/ultralytics/yolov3 && cd yolov3 to get the codebase.
2
Install Dependencies
Install required packages with pip install -r requirements.txt ensuring Python >=3.8 and PyTorch >=1.8 are installed.
3
Load Pretrained Model
Use PyTorch Hub or from ultralytics import YOLO; model = YOLO("yolov3u.pt") to load the pretrained model for inference.
4
Train Model
Train on custom datasets using model.train(data="coco8.yaml", epochs=100, imgsz=640).
5
Perform Inference or Export
Run inference or export the model to ONNX or TFLite formats for deployment.
📊

Strategic Context for YOLOv3

Get weekly analysis on market dynamics, competitive positioning, and implementation ROI frameworks with AI Intelligence briefings.

Try Intelligence Free →
7 days free · No credit card
Pricing
Model: open-source

YOLOv3 is open-source and free to use. Enterprise licensing is available through Ultralytics for commercial applications.

Assessment
Strengths
  • Supports multiple export formats (ONNX, CoreML, TFLite) for flexible deployment.
  • Compatible with YOLOv5 updates ensuring ongoing maintenance.
  • Achieves 28.2 mAP on COCO dataset at 22 ms per 320×320 frame.
  • Straightforward training process via Ultralytics API on custom data.
  • Available in multiple implementations including PyTorch and TensorFlow.
Limitations
  • Original Darknet repository is inactive, relying on community forks like Ultralytics for updates.
  • Documentation specific to YOLOv3 is limited; users often refer to newer YOLO versions for general principles.
  • Original Darknet .cfg model archives are no longer maintained.