Introduction to Deep Learning with DeepDrivePL Image Classification course

12 Modules, 11 Homeworks

1. Introduction
   • Image representation, color spaces, convolution, morphological operations
   • Preprocessing, Datasets, data visualizations
2. First CNN training
   • Loss function, linear regression vs classification problem
   • Metrics, Backrpopagation
3. Neural network construction
   • Activation functions, optimizers
   • Deep double descent
   • LR strategies, Pooling, dropout, normalization
4. Transfer learning
   • Modern image classification NN
   • Training from scratch vs pretrained model, fine tuning
5. Data agumentation
   • Augmentation types, geometrical and arithmetical transformations
   • Kornia/Albumentations augmentation
6. Hyperparameters optimization
   • HPO vs NAS
   • Optuna, Parallel coordinates
7. Analysis of model output
   • Confusion matrix, layers visualization
   • Occlusion sensitivity, T-SNE, UMAP
   • SmoothGrad, GradCam
8. Self superviced learning
   • Rotation prediction
   • SESEMI, SimCLR, BYOL
9. Binary classification
   • Binary vs Multiclass vs Multilabel vs MultiTask
   • Sigmoid vs Softmax
   • Precision & Recall, Balanced Accuracy, PR Curve, ROC, F1
10. Model optimization
    • AMP, FP32, INT8
    • Pytorch model quantization
    • Pruning, TensorRT
    • Knowledge distillation
11. Model deployment
    • Gradio, Torch Serve, fastapi
    • Docker
    • Nvidia Triton