search

AayushKrChaudhary / RITnet

This is a winning model of OpenEDS Semantic Segmentation Challenge
MIT License
72 stars 28 forks source link

Testing best_model on openEDS #8

Closed AbdouMechraoui closed 2 years ago

AbdouMechraoui commented 2 years ago

Hi there, Thank you so much for making the code available. I tried to test best_model on OpenEDS data, but the results don't look satisfying. I wanted to ask, what dataset was best_model trained on? Also, how well does best_model generalize to real-world data? I plan to use your model for eye segmentation of real-world images, and I was wondering if you can give me an insight on how well it performs, or what to look out for. Below I have attached some of the results I got when I tested on OpenEDS. Is there any way to improve the resulting segmentations?

0000 0006

AayushKrChaudhary commented 2 years ago

The images are not from OpenEDS-2019. The best model is trained on real images of OpenEDS-2019. Retraining the model in this dataset will help improve the result.