Udacity SDC Nano degree

Semantic Segmentation Project

Acknowledgement

In addition to the code walkthrough I have refereed to following two works during the development

• Repo 1
• Repo 2

Brief overview

A fully convolutional neural network (FCN) based on VGG-16 is used to perform segmentation. The network learns to identify drivable area. The training and testing are performed on KITTI dataset.

Approach

Architecture

A pre-trained VGG-16 network was converted to a fully convolutional network by converting the final fully connected layer to a 1x1 convolution and setting the depth equal to the number of desired classes (in this case, two: road and not-road). Performance is improved through the use of skip connections, performing 1x1 convolutions on previous VGG layers (in this case, layers 3 and 4) and adding them element-wise to upsampled (through transposed convolution) lower-level layers (i.e. the 1x1-convolved layer 7 is upsampled before being added to the 1x1-convolved layer 4). Each convolution and transpose convolution layer includes a kernel initializer and regularizer

Optimizer

The loss function for the network is cross-entropy, and an Adam optimizer is used.

Training

Cross-entropy based Adam optimizer is used. The hyperparameters used for training are:

• keep_prob: 0.7
• learning_rate: 0.0005
• epochs: 100
• batch_size: 10

Average loss per batch during different ephochs

1. Epoch 1 : 0.63
2. Epoch 50 : 0.18
3. Epoch 90 : 0.009
4. Epoch 100 : 0.008

Segmentation on test images

Results are not good when there is too much shadow (Need to add such examples)

Original README

Semantic Segmentation

Introduction

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

Setup

GPU

main.py will check to make sure you are using GPU - if you don't have a GPU on your system, you can use AWS or another cloud computing platform.

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.

Start

Implement

Implement the code in the main.py module indicated by the "TODO" comments. The comments indicated with "OPTIONAL" tag are not required to complete.

Run

Run the following command to run the project:

python main.py

Note If running this in Jupyter Notebook system messages, such as those regarding test status, may appear in the terminal rather than the notebook.

Submission

1. Ensure you've passed all the unit tests.
2. Ensure you pass all points on the rubric.
3. Submit the following in a zip file.

• helper.py
• main.py
• project_tests.py
• Newest inference images from runs folder (all images from the most recent run)

Tips

• The link for the frozen VGG16 model is hardcoded into helper.py. The model can be found here
• The model is not vanilla VGG16, but a fully convolutional version, which already contains the 1x1 convolutions to replace the fully connected layers. Please see this forum post for more information. A summary of additional points, follow.
• The original FCN-8s was trained in stages. The authors later uploaded a version that was trained all at once to their GitHub repo. The version in the GitHub repo has one important difference: The outputs of pooling layers 3 and 4 are scaled before they are fed into the 1x1 convolutions. As a result, some students have found that the model learns much better with the scaling layers included. The model may not converge substantially faster, but may reach a higher IoU and accuracy.
• When adding l2-regularization, setting a regularizer in the arguments of the tf.layers is not enough. Regularization loss terms must be manually added to your loss function. otherwise regularization is not implemented.

Using GitHub and Creating Effective READMEs

If you are unfamiliar with GitHub , Udacity has a brief GitHub tutorial to get you started. Udacity also provides a more detailed free course on git and GitHub.

To learn about REAMDE files and Markdown, Udacity provides a free course on READMEs, as well.

GitHub also provides a tutorial about creating Markdown files.