makarandmandolkar/carnd-semantic-segmentation

Semantic Segmentation

Introduction

In this project, you'll label the pixels of a road in images using a Fully Convolutional Network (FCN).

Figure: Results after 10 epochs of training with augmentation (Command to reproduce: python main.py 10 16 0.5)

Check road segmentation with 4k video on youtube:

Setup

Frameworks and Packages

Make sure you have the following is installed:

• Python 3
• TensorFlow
• NumPy
• SciPy

Dataset

Download the Kitti Road dataset from here. Extract the dataset in the data folder. This will create the folder data_road with all the training a test images.

4K video data

wget https://drive.google.com/open?id=0B_6iW8KaJFXOQmhaWU56dlBDY28

Training

Run the following command to run the project:

python main.py

Parameters:

• number of epoch
• batch size
• keep probability

#python main.py  EPOCHS  BATCH_SIZE  DROPOUT
python main.py 10 16 0.5

Figure: Training loss after 10 epochs of training with augmentation (Command to reproduce: python main.py 10 16 0.5)

Data augmentation

get_batches_fn() in helper.py

• flipping every image
• translating +/- 100 in x axis and +/- 30 in y axis
• brightness +/- 150 (in 0-255 range)

Example batch: