This repository contains image classification task program for carvana car masking data set found in Kaggle (https://www.kaggle.com/c/carvana-image-masking-challenge)

The data is converted to 512 x 512 images for both image and masks. So, please create a directory structure like below-

| project directory
| train512
|-- images
|---- 0ce66b539f52_04.jpg
|---- ......
|-- labels
|---- 0ce66b539f52_04.png
|---- ......
| validation512
|-- images
|---- 0ce66b539f52_10.jpg
|---- ......
|-- labels
|---- 0ce66b539f52_10.png
|---- ......
| results
| segmentation
| README.md

For this task, UNET model was used, the trained model weights can be found here and sample real image and predictions are kept inside results/ directory.

train.py contains training program

prediction.py is a command line program which can show prediction for a given input image and also shows prediction performance upon provided label/mask image.

Sample image input

Label image

Corresponding Prediction