Final Assignment: Cityscape Challenge

Welcome to the repository for 5LSM0: Final Assignment: Cityscape Challenge, the final project for the course 5LSM0: Neural Networks for Computer Vision, offered by the Department of Electrical Engineering at Eindhoven University of Technology. This course is hosted by the Video Coding & Architectures research group.

📦 Required Libraries & Environment Setup

This project uses a pre-configured Docker container with all necessary dependencies:

• Dataset: Cityscapes
• Docker Image: docker://tjmjaspers/nncv2025:v7
• Singularity Container: container.sif

🔧 Installation Instructions

1. Clone this repository:

   git clone <repo-url>
   cd <repo-directory>

2. Download Docker and Cityscapes dataset: Run the following script to download both:

   sbatch download_docker_and_data.sh

3. Optional: To change the Docker version, edit the download_docker_and_data.sh script accordingly.

💡 All necessary modules are included in the container.sif Docker image. No manual pip installations are required.

---

🚀 How to Train the Model

1. Configure the Training Script:
   In main.sh, replace the script reference with the desired training configuration (e.g., train_Segformer_Finetuned.py). All training scripts are prefixed with train_.

2. Adjust Hyperparameters:
   Modify learning rate, batch size, and number of epochs directly in main.sh.

3. Set Training Time and GPU Settings:
   In jobscript_slurm.sh, set the maximum training time and (optionally) specify the number of GPUs.

   ⚠️ If you use multiple GPUs, ensure the training script supports parallel training (manual changes may be required).

4. Submit the Job to the Supercomputer:
   Use the following command to submit the job via SLURM:

   sbatch jobscript_slurm.sh

   To check the job status in the queue:

   squeue

---

🧪 How to Test the Model on Codalab

If you have a trained model checkpoint and want to participate in the Codalab evaluation:

1. Select Your Model File:
   Find the appropriate model script (files prefixed with Model_) and rename it to model.py.

2. Update Pre/Post-Processing:
   Modify process_data.py with the correct input size and any desired preprocessing.

3. Prepare the Checkpoint:
   Rename your .pth file to model.pth.

4. Package for Submission:
   Zip the following files:

   • model.py
   • model.pth
   • process_data.py

   Submit the zip file to the Codalab competition platform.

---

👤 Author & Codalab Info

• Name: Shao-Ruei Huang
• TU/e Email: s.huang5@student.tue.nl
• Codalab Username: TUe-Ray