Semantic_Segmentation_project

Title: Efficient Domain Adaptation for Real-Time Semantic Segmentation with Lightweight Networks and Discriminators

For "Advanced Machine Learning" course at Politecnico di Torino

Made by:

• Ivan Magistro Contenta
• Yalda Sadat Mobargha
• Luca Sturaro

The repository contains:

• model/: definitions of models and trained models
  • STDC-net: model_stages.py and stdcnet.py
  • BiSeNet v1: bisenetv1.py
  • best_models/ contains trained models on Domain Shift and Domain Adaptation tasks
    • Domain Shift
      • without data augmentation: p2c_lr_0001_bs_8_notaug_Saved_model_epoch_50.pth
      • with data augmentation: p2c_lr_0001_bs_8_aug_Best_model_epoch_35.pth
    • Domain Adaptation
      • STDC-net: p3_lr_0001_lrD_00001_bs_8_Saved_model_epoch_50.pth
      • BiSeNet v1: p4_bisenetv1_domadpt_lightdiscr_BSv1_Best_model_epoch_25.pth
• notebook_files/: implementation of training, validation and other techniques
  • run_stdc_bisenetv1.ipynb: useful to run different tasks of the code on GPU (Colab)
  • cpu_execution.ipynb: useful to run different tasks of the code on CPU (Colab)
  • metrics.ipynb: it contains the metrics of different networks and discriminators, but also the outputs of the best models to be compared to the images and ground truths