Open Lanme opened 4 years ago
It's expected in unet-like segmentation architectures. Instance segmentation models are better in predicting masks, if the object you are applying segmentation is an instance (as in this case).
You can try this repo if you are looking for an instance segmentation model for id card detection: https://github.com/fcakyon/id-card-detector.
Or you can directly install it by pip install id-card-detector
.
Test some images from midv datasets with the provided model,the above seems difficult to identify accurately,can someone explain?