Optic Disc/Cup Segmentation and Glaucoma Classification with FCNs Zifeng Wu, Chunhua Shen, Anton van den Hengel and Jianpeng Zhang AIML
Optic Disc/Cup Segmentation Cup Disc
Inference Pipeline Classify (predict & argmax) Location Score maps Model A Location Model B1 Rescale & average Crop & Predict rescale Local Argmax Pad normalization Model B2
Backbone Networks • ResNet-50 [1] • ResNet-101 [1] • Wide ResNet-38 [2] [1] Deep residual learning for image recognition, K. He, X. Zhang, S. Ren and J. Sun, CVPR, 2016. [2] Wider or deeper: Revisiting the ResNet Model for visual recognition, Z. Wu, C. Shen and A. van den Hengel.
FCN Details • DeepLab-like Dilated (prediction stride: 32 -> 8) • U-net-like up-sampling (prediction stride: 8 -> 4) • Jump connections from high-resolution feature maps • Online hard example mining
Glaucoma Classification Glaucoma? Yes/No?
Inference Pipeline Model C1 Model C2 Classify (predict & argmax) Average & argmax Local Predict normalization Model A Location Model C3 Crop Model C4
Thanks!
Recommend
More recommend