TDT17-2023

VI-tasks on Cityscapes

Expected folder structure

data/cityscapes
├── gtFine
|   ├── test
|   ├── train
|   ├── val
├── leftImg8bit
|   ├── test
|   ├── train
|   ├── val

Tasks

• Select a network structure
• Train a network for image segmentation on cityscapes
• Perform semantic segmentation (inference) on cityscapes
• Run inference on 3 random images from Trondheim

Sources

• CityScapes - large-scale dataset that focus on semantic understanding of urban street scenes
• DeepLabV3+ - Encoder-Decoder with Atrous Separable Convolution for Semantic Image Segmentation
• Timm - PyTorch Image Models
• segmentation_models.pytorch - Segmentation models with pretrained backbones
• Pytorch Dataset
• TorchMetrics - PyTorch metrics implementations and an easy-to-use API to create custom metrics
  • Accuracy
  • Jaccard Index - intersetion over union or jaccard similarity coefficient
  • F-Beta Score
• PyTorch lightning