A Comprehensive Study on Colorectal Polyp Segmentation with ResUNet++, Conditional Random Field and Test-Time Augmentation

Benchmark Model Rank Results
medical-image-segmentation-on-cvc-videoclinicdbResUNet++ + TTA#1Dice: 0.8125mIoU: 0.8467Recall: 0.6896precision: 0.6421
medical-image-segmentation-on-cvc-videoclinicdbResUNet++ + TTA + CRF#2Dice: 0.8130mIoU: 0.8477Recall: 0.6875precision: 0.6276
medical-image-segmentation-on-cvc-videoclinicdbResUNet++ + CRF#4Dice: 0.8811mIoU: 0.8739Recall: 0.7743precision: 0.6706
medical-image-segmentation-on-cvc-colondbResUNet++ + TTA#7mean Dice: 0.8474mIoU: 0.8466
medical-image-segmentation-on-etis-laribpolypdbResUNet++ + TTA#22mean Dice: 0.6136mIoU: 0.7458
medical-image-segmentation-on-cvc-clinicdbResUNet++ + TTA#40mean Dice: 0.9020
medical-image-segmentation-on-cvc-clinicdbResUNet++ + CRF+ TTA#41mean Dice: 0.9017
medical-image-segmentation-on-kvasir-segResUNet++ + TTA + CRF#48mean Dice: 0.8508mIoU: 0.7800FPS: 69.59