Advanced Deep Learning project
Labeling pixels of a road in images using a FCN using Semantic Segmentation approach based on VGG Model.
To construct a fully convolutional neural network based on the VGG-16 Image Classifier architecture for performing semantic segmentation to identify drivable road area using the KITTI dataset.
Architecture
The architecture used to build the FCN is known as FCN-8. FCNs are known to work really well on segmentation tasks. A pre-trained VGG-16 network was converted to a FCN by changing the final FC layer to a 1x1 convolution with depth equal to the number of desired classes.(We have twop classes: Roads and Non-Roads) Performing 1x1 convolutions on VGG Layers 3 and 4, and adding them element-wise to the initially upsampled transposed convolution lower-level layers (initially 1x1-convolved layer 7 is upsampled before getting added to the 1x1-convolved layer 4). We are using Kernel Intializers and Regularizers for each convolution and transpose convolution layer.
Hyper parameters
- Learning rate = 0.0001
- Batch size = 8
- Epochs = 20
- Dropout = 0.5
- Initializer standard deviation = 0.01
Optimizer
The loss function for the network is cross-entropy, and an Adam optimizer is used.
Results
Loss per batch tends to average below 0.200 after two epochs and below 0.100 after ten epochs. Average loss per batch at epoch 20: 0.048, at epoch 30: 0.074, at epoch 40: 0.036, and at epoch 50: 0.030.
Environment Setup
Alternatively you can run:
pip install requirement.txt
Downloads
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.
Download the CityScapes dataset
KITTI Dataset
CityScape Dataset
