The first time for me to construct a custom dataset and apply it in the mmdection.

The whole pipeline is as follows:

I. Collect and prepare the raw dataset.

II. Convert the raw dataset into COCO style.

III. Add and movify the mask-rcnn configuration in mmdection API to fit our requirement.

IV. Train the model by mmdection.

V. Evaluate and visualize the test dataset.

VI. Modify the result into mask image so that we can insert it smoothly in our pipeline.

Read more