Table 3 Specifications of the proposed and alternative approaches.

From: Cataract-1K Dataset for Deep-Learning-Assisted Analysis of Cataract Surgery Videos

Model

Backbone

Params.

Upsampling

Target

Reference

DeepPyramid

VGG16

33.57 M

Bilinear

Medical Images

22

Adapt-Net

VGG16

24.69 M

Bilinear

Medical Images

17

UNet++

VGG16

24.24 M

Bilinear

Medical Images

33

ReCal-Net

VGG16

22.93 M

Bilinear

Medical Images

21

CPFNet

VGG16 | ResNet34

39.17 M | 34.66 M

Bilinear

Medical Images

34

CE-Net

VGG16 | ResNet34

33.50 M | 29.90 M

Trans Conv

Medical Images

35

FED-Net

ResNet50

59.52 M

Trans Conv & PixelShuffle

Liver Lesion

36

scSENet

VGG16 | ResNet34

22.90 M | 25.25 M

Bilinear

Medical Images

37

DeepLabV3+

ResNet50

26.68 M

Bilinear

Scene

38

UPerNet

ResNet50

51.26 M

Bilinear

Scene

39

U-Net+1

VGG16

22.55 M

Bilinear

Medical Images

40

  1. In “Upsampling” column, “Trans Conv” stands for Transposed Convolution.
  2. 1Note that UNet + is an improved version of UNet, where we use VGG16 as the backbone network and double convolutional blocks (two consecutive convolutions followed by batch normalization and ReLU layers) as decoder modules.
Back to article page