How to improve the quality of segmentation of converted model

[greg007-web]

I made an adaptation to take photos and perform image segmentation. When I convert the original model, I lose segmentation quality; it segments outside the object, which affects my work with measurements. Is there anything I can do to improve this quality in the code or during the conversion? Thank you.

Thanks.

[Digital2Slave]

1. refine your custom dataset label without missing.
2. fine tuning your segment model with your custom dataset.
3. set prob_threshold and nms_threshold values according to your measurements.