This repository presents a ROI calibration/alignment workflow for industrial inspection, involving deep learning-based keypoint detection.
The use case presented is for fashion industry, in particular for trousers.
A keypoint detection model is trained with a subset of DeepFashion2 dataset, which contains images and annotations with "trousers" category.
The definition of keypoints is as below:
Model weight:
By computing the barycentric coordinates of reference ROIs (bounding boxes) corresponding to the triangles drawn from the keypoints detected, ROIs can be auto aligned and adjusted on target images.
Robust: The position, size, and ratio of ROIs are adjusted accordingly based on keypoints.
The open-source / permissive license applied to the code only, for demonstrating the keypoint detection results can be used for ROI calibration.
Refer DeepFashion2 for its license (by this time, strictly for acedemic research only, not for commercial use).
You should use licensed or self-annotated dataset for any commercial usage.
