Skip to content

AmanjotBhullar/Transfer_Learning_Image_Classification

Repository files navigation

Transfer Learning with Inception-v3 for Histopathological Image Classification

The classify.py, convert-model.py, label.py, predict-request.py, retrain.py files were taken from the TensorFLow transfer learning api -- https://www.tensorflow.org/hub/tutorials/image_retraining.

The BreakHis dataset was used for the image classification, and available here -- https://web.inf.ufpr.br/vri/databases/breast-cancer-histopathological-database-breakhis/. Download and move dataset to this repository.

File summaries

'1. Dataset_Maker.ipynb':

The BreakHis dataset contains histopathology images in 40x, 100x, 200x, 400x magnifications. This file creates seperate directories for each magnification in the tf_files directory. This is done because an image classifier will be created using the 40x, 100x, 200x, and 400x datasets. The accuracies of each of the 4 classifiers is compared.

'2. Transfer_Learning.ipynb':

Transfer learning is done on the last layer of Inception-v3, a pre-trained convolutional neural netowrk, using each of the 4 magnifications.

'3. Data_Analysis.ipynb':

The accuracies of the classifiers are compared.

About

Transfer learning on Inception-v3 using the BreakHis dataset.

Topics

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published