Skin Cancer Detection Android App(Image Processing)
Here are the Codes, Datasets and the Android App required to create the Skin Cancer Detection App.
Data
The dataset is orignally available at https://dataverse.harvard.edu/dataset.xhtml?persistentId=doi:10.7910/DVN/DBW86T . follow the steps given on https://github.com/syed-hamza/Skin-Cancer-Android-App/blob/master/Data.md for dataset prepartion.
Structure
The data should be arranged in the following struture.
https://github.com/syed-hamza/Skin-Cancer-Android-App/blob/master/input.md
Code
- Inception-V3 - https://github.com/syed-hamza/Skin-Cancer-Android-App/blob/master/src/SkinCancer_inceptionModel.ipynb
- nasnet - https://github.com/syed-hamza/Skin-Cancer-Android-App/blob/master/src/SkinCancer_nasnetmobile.ipynb
Inference
The given code uses inception_v3 at https://keras.io/applications/#inceptionv3.Inception model gave me a slightly better result than nasnetmobile and training the model was comparitively quick.
Inception(Validation_Accuracy):
validation accuracy:85.76%
NasNetMobile(Validation_Accuracy):
validation accuracy:84.72%
From this we can conclude that eventhough nasnet fluctuates less than inception inception provides a better accuracy,hence is better suited for the app.
Regards
I am a student in highschool and this is my first project using Deep Learning. The following project took me several weekends along 2 months. I hope it is helpful to everyone reading this. All suggetions are welcome.