🚧 This is still under development 🏗

Semantic segmentation has become an important component of self-driving vehicles. It allows the car to understand the surroundings by classifying every pixel of the input image.

To run inference on the pre-trained models, please use segmentor.py.

from segmentor import Segmentor
seg = Segmentor()
cls_output, img_viz = seg.semantic_segmentation(image=image, visualization=True)

cls_output is the pixel-wise classification result for all the categories. img_viz is a RGB image generated based on cls_output

The best way to run some actual test is using test.py. You must specify the image path by changing the path variable.

The pre-trained weights are stored in the ./weights directory.

train.py is the training script. utils.py contains all the categories (classes). You can modify them based on your dataset.

training doesn't work very well. I trying to fix the issue.

The goal is to achieve real time semantic segmentation. Therefore, choosing & developing the appropriate model is critical. I experimented with UNet and ENet, but realized drawbacks to both of these architectures. I eventually landed on ICNet, which stands for image cascade network

Here is a simple benchmark comparison between ICNet and other popular semantic segmentation models. These images and visualizations are from the original ICNet paper, which can be found here.

This project is created for the self-driving golf cart project that I have been working on. For more information on that, please refer to the Github page, or my website.

If you have questions, comments or concerns, please contact me at contact@neilnie.com.

./weights/new-enet-5.h5: 512x512x26

./weights/enet-c-v1-2.h5: 640x320x16

./weights/enet-c-v2-1.h5: 1024x512x16

The channel value correspond to the number of classes.