This project demonstrates the use of a pretrained ResNet50 model for semantic image segmentation.
The goal is to identify and separate different objects or regions within an image by applying a segmentation model.
The dataset consis of real images and their respective ground truth masks.
Each image can be segmented into at most 9 different regions
Example image
It utilizes a ResNet50 model pre-trained on ImageNet and fine-tuned for semantic segmentation using a U-net.
The model is tested on a dataset of images, and segmentation masks are predicted to classify different regions of the images.
Please refer to report.pdf for detailed results and analysis
Some sample model predictions
