search

ESandML / FCBFormer

Official code repository for: FCN-Transformer Feature Fusion for Polyp Segmention (MIUA 2022 paper)
Apache License 2.0
67 stars 22 forks source link

Is this model available for multiple classes? #5

Open wonchul-kim opened 1 year ago

wonchul-kim commented 1 year ago

Is this model available for multiple classes?

ESandML commented 1 year ago

Hi, the model is currently only available for binary classes. If you would like to modify the code to run for multiple classes, you would just need to adjust the number of output channels to the number of classes and change the loss function. The loss function is currently the sum of the Dice loss and the BCE loss, which are both intended for the binary case, so you would need to change this to a loss intended for the multi-class case, e.g. CE loss.

wonchul-kim commented 1 year ago

Is there any benchmark for the performance of multiple classes dataset?

ESandML commented 1 year ago

We have not produced results for any multi-class image segmentation benchmark. However, the EndoScene dataset is a good example of such a benchmark for colonoscopy images, and COCO and CityScapes are popular more generally.