This work involves the use of an encoder decoder architecture CNN for semantic segmentation of the image. We took inspiration from LinkNet for the same and trained our model oon both the mapillary dataset and the Berkley Deep Drive dataset.
https://drive.google.com/file/d/1WFPjmI9Tx_5_-UgVa2A2-6RhbGRrl2Zt/view?usp=sharing
For training or testing please keep all your RGB images and their corresponding ground truths in separate folders but bearing the same name. The input image must be a jpg and the ground truth must be a greyscale png with each pixel value being its anootated class index value.
For training use the following
foo@bar$ python3 all_linknet.py -m <any of the legal modes>
The checkpoint will be stored in the ./model_files directory. Feel free to train it yourself or you could use one of our checkpoints for the same.