Zkouška 2025

1. Deep Learning

• Visual Transformers, architecture (draw scheme)
• Explain self attention mechanism, use appropriate math formulas
• How are ViT better then CNN

2. Image Matching

• Describe SIFT algorithm, possible parametres
• What are the pros/cons of local feature orientation
• How to predict orientation using NN. Loss? What is the source of GT
• How to take images of the same view so SIFT fails

3. RANSAC

• Algorithm, parametres
• Derive stopping criterium, what does it guarantee
• Faster and more precise versions of RANSAC
• How to use RANSAC to improve image retrieval

4. Image Retrieval

• image retrieval task formulation
• recall@k
• average precision
• how to change recall@k so it can be used as loss