Dataset Export: YOLOv6 Model Training

Roboflow provides the option to export object detection datasets with annotation files compatible with YOLOv6.

Written by Mohamed Traore

Last published at: July 1st, 2022

MT-YOLOv6, or "YOLOv6," is a single-stage object detection framework dedicated to industrial applications, with hardware-friendly efficient design and high performance.


The new object detection model is from 美团 (Meituan Dianping).

Benchmarking YOLOv6 vs. Other Single-Stage Object Detection ModelsYOLOv6 benchmarking

Background:

YOLOv6-nano achieves 35.0 mAP on COCO val2017 dataset with 1242 FPS on T4 using TensorRT FP16 for bs32 inference, and YOLOv6-s achieves 43.1 mAP on COCO val2017 dataset with 520 FPS on T4 using TensorRT FP16 for bs32 inference.

YOLOv6 is composed of the following methods:

  1. Hardware-friendly Design for Backbone and Neck
  2. Efficient Decoupled Head with SIoU Loss

(Source)

Exporting Your Images to Train YOLOv6

Roboflow is home to over 30 options for annotation format imports and exports so you can quickly leverage multiple model architectures for custom model training from our model zoo. When exporting your dataset with Roboflow, you can rest assured that the images and annotation files will be in the correct format for you to start training your model, right away!

Roboflow's import and export format options also provide the ability to convert annotation formats, bring in annotation files from other platforms, and still enjoy Roboflow's AutoML for easy deployment solutions and model-assisted labeling, without having to re-annotate all of your images!

After starting your project‍, and successfully generating a dataset version for your object detection project, you're ready to export your images and annotation files, and train a custom YOLOv6 model!

Selecting the YOLOv6 Export FormatSelecting the YOLOv6 Export Format

Training Your Model