This is one of the best semantic segmentation model, works pretty well on the Camvid dataset with 11 classes

CUDA_VISIBLE_DEVICES=0 python train_segmentation.py --model espnetv2 --s 2.0 --dataset sample --data_path ~/EdgeNet/vision_datasets/sample_dataset/ --batch-size 1 --crop_size 512 256 --model espnetv2 --s 1.5 --lr 0.009 --scheduler hybrid --clr-max 61 --epochs 100

• Input: vision_datasets/sample_dataset/images
• Target: vision_datasets/sample_dataset/annotations
• First run the convert_to_gray.py file which will convert all the annotations into grayscale images
• Then add the image names and respective grayscale annotations pair in vision_datasets/sample_dataset/train.txt file
• Do the same for val.txt file
• Start running the training process

1. python test_seg.py

This file run on the live webcam feed

2. python tester.py

This runs on the images on the images in the input folder

The original work is done by Sachin Mehta, the repository can be found here