In this project, you'll label the pixels of a road in images using a Fully Convolutional Network (FCN).
This project implemented FCN-8s-VGG16 proposed in the paper Fully Convolutional Networks for Semantic Segmentation using Tensorflow.
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.
Make sure you have the following is installed:
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 and test images.
Run the following command to run the project:
python main.py
This will train the model (it will check ./model/model.ckpt
to pick up the training, if no file it will train from scratch), save model as model.ckpt
and frozen .pb
file, and process all the test images in test folder.
Note If running this in Jupyter Notebook system messages, such as those regarding test status, may appear in the terminal rather than the notebook.
When the project is run, the frozen VGG16
model will be downloaded. The model can be found here, too.
-
Running the script
python vis_tfboard.py
will generate a foldervgg16logdir
which you can exploit usingtensorboard --logdir="vgg16logdir"
. Go to the address designated using web browser to see the graph and Operation nodes. Look forlayer3_out
,layer4_out
andlayer7_out
. You will understand why we want the outputs of these nodes. It's very clear that for the 3 dense layers, two of them are converted to the fully convolutional layers and the final layer is removed. -
Load the graph and list all the Operations and the outputs attached to them by running the script
python explore_vgg.py
. Doing this is more difficult for you to find which outputs of which layers you want to capture. Too many nodes in VGG16.
- The link for the frozen
VGG16
model is hardcoded intohelper.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.