Project 1 Repository for the course Deep Learning in Computer Vision

• (optional for HPC): module load python3/3.x.x
• python3 -m venv venv_1
• source venv_1/bin/activate
• make requirements

• make train

• A description of your architecture and how you designed it? Did you start out with something else? How/why did you decide to change it?
• How did you train it? Which optimizer did you use? Did you compare using different optimizers? Did you do other things to improve your training?
• Did you use any data augmentation? Did you check if the data augmentation improved performance?
• Did you use batch normalization? Does it improve the performance?
• What is the accuracy of your network? Which test images are classified wrong? Any of them for obvious reasons?
• Did you use transfer learning? Does it improve the performance?
• Compute the smoothgrad saliency map and plot it for some example images. Make sure you can you explain how adding Gaussian noise to an image is equivalent to drawing a sample from a normal distribution centered at that image.
• Did you use ChatGPT or similar tools? If yes, please briefly describe how you used it and how they were useful.