Blog
Blog
No items found.
YOLOv3 is an open-source state-of-the-art image detection model. You will find it useful to detect your custom objects. Roboflow provides implementations in both Pytorch and Keras. YOLOv3 has relatively speedy inference times with it taking roughly 30ms per inference. It takes around 270 megabytes to store the approximately 65 million parameter model. There are also variations within YOLOv3 such as Tiny-YOLOv3 which can be used on Rasberry Pi.
YOLOv3 made the initial contribution of framing the object detection problem as a two-step problem to first identify a bounding box (regression problem) and then identify that object's class (classification problem).
Image in Courtesy of Ethan Yanjia Li
YOLOv3 is an incredibly fast model with it having inference speeds 100-1000x faster than R-CNN. When it was released, YOLOv3 was compared to models like RetinaNet-50 and Retina-Net-101. It had a state-of-the-art performance on the COCO dataset relative to the model's detection speed and inference time, and model size. Below are some of the results comparing YOLOv3 to models of the time.
We find YOLOv3 to have slightly poorer performance than EfficientDet, Scaled YOLOv4, and other modern models on an example custom dataset.
Training a YOLOv3 Object Detection Model with a Custom Dataset: https://blog.roboflow.com/training-a-yolov3-object-detection-model-with-a-custom-dataset/
YOLOv3 PyTorch
is licensed under a
GPL-3.0
license.
You can use Roboflow Inference to deploy a
YOLOv3 PyTorch
API on your hardware. You can deploy the model on CPU (i.e. Raspberry Pi, AI PCs) and GPU devices (i.e. NVIDIA Jetson, NVIDIA T4).
Below are instructions on how to deploy your own model API.
