Lyft

Udacity-Lyft road-car semantic segmentation for Carla simulator

to compete with other udacity students for Udacity-Lyft Road-Car Carla simulator semantic segmentation !

Setup

github

https://github.com/byronrwth/Lyft.git

Frameworks and Packages

my Anaconda environment requirements: requirements_OD-lab-py35.txt

Dataset

multiple dataset for training from Carla simulator

JWT token

need for a token to submit

Start

Implement

Implement the code in the /home/workspace/Example/lyft_multi.py

Run

Run the following command to run the project:

cd /home/workspace
python Example/lyft_multi.py /Example/test_video.mp4

Submission

tester 'Example/lyft_multi.py' grader 'Example/lyft_multi.py' submit

Tips

• The link for the frozen VGG16 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.

-- use batch inference to increase infere speed from 3.8 FPS to 9FPS, with inference batch size = 16

-- crop origin image (600, 800) to (400, 800) and resize to (256,512) for VGG input size, remember decrop back to (600, 800) before genereate final raod binary array and car binary array for score