CheXpert

ABOUT:

The CheXpert dataset contains Xrays for 14 different Feature Classes, the goal is to identify each label(i.e. feature class) from just passing the Xray image as an Input.

The Size of the dataset is 11GB and contains 2,23,414 imaegs for training and 234 images for validation , as the size is quite large it takes a lot of time to compute. Hence only 10% of the dataset was used in this project.

The (10%)dataset was then split into train and test with 22,341 and 500 images respectively.

(Nvidia's GTX 1060 6GB was used for training)

ROC

Visualizing feature maps for Transition layer 1

Outputs:

1. For Class Support Devices

2. For Class No Finding