Train YOLOv7 Segmentation on Custom Data 🤔

Image segmentation is one of the major problems highlighted on the research side. There are many good algorithms available for image segmentation, i.e. Mask-RCNN, mm-detection, etc, and each of them has its usage and advantage.

YOLOv7 is the latest object detection algorithm in terms of accuracy as compared to other YOLO variants which include, YOLOv3, YOLOv4, YOLOv5, etc. If you want to train YOLOv7 on custom data, you can check my article “YOLOv7 Training on Custom Data”

A few days ago, YOLOv7 released the image segmentation module, but not as a primary module. The data preparation and usage are based on YOLOv5, although the algorithm is interlinked with the original YOLOv7 object detection weights.

Fig-1.1: YOLOv7 Instance Segmentation

Before moving toward implementation, I must mention SparkIntelligence, which supports bringing this article to an audience like you.

⭐ follow the SparkIntelligence LinkedIn page and website to get regular updates about computer vision and embedded vision. SparkIntelligence has worked on different products i.e. Fork Vision, People Counting, Personal Protective Equipment (PPE) Detection, Object Blurring Tool, Warehouse Automation, and Eye Disease Detection. SparkIntelligence team includes Ph.D. and Bachelor’s graduates, who have experience in purely computer vision domain.

In this article, I will discuss the mentioned modules.

• Label Custom Data for Segmentation on Roboflow.
• Train YOLOv7 Segmentation algorithm on Custom Data.

Label Custom data for Segmentation on Roboflow

There are many labeling tools, i.e. labelImg, label me, CVAT, label box, roboflow, etc. In this article, I will go with roboflow as it has a very good structure and an easy labeling tool for work without any installation issues.

The data labeling for segmentation will be a polygon box, while data labeling for object detection will be a bounding box

The steps for data labeling are as follows,

1- Creation of Workspace
2- Custom Option to select
3- Create Private Project
4- Upload your dataset
5- start labeling