This repository contains homework projects developed as a part of Deepdrive image classification course.
Click on a chapter number to go to the notebooks and results for a given chapter.
| Chapter | Task | Main libraries |
|---|---|---|
| 02 | Perform basic data visualization of MNIST-like dataset | numpy |
| 03 | Train very simple CNN model for classification of MNIST-like dataset | PyTorch |
| 04 | Apply various regularization techniques to improve model from the previous chapter | PyTorch Ligthning (PL), Weights & Biases (W&B) |
| 05 | Training from scratch vs Transfer Learning for satellite image classification on RESISC45 dataset | PyTorch Image Models (TIMM), PL, W&B |
| 06 | Improve models from previous chapter using data augmentation | Albumentations, PL, TIMM, W&B |
| 07 | Run hyperparamter optimization (e.g. Optuna) on models from the two previous chapters | Optuna, PL, TIMM, W&B |
| 08 | Interpretability analysis (e.g. occlusion sensitivity and GradCAM) for models from chapters 05-07 | Captum, PL, TIMM, W&B |
| 09 | Run Self-Supervised Learning (SSL) on unlabeled dataset as a pretraining for supervised model | Lightly, PL, TIMM, W&B |
| 10 | Binary classification with imbalanced dataset (incorporating weighted loss and balanced accuracy) | PL, TIMM, W&B, FiftyOne |
| 11 | Model optimization (e.g. pruning or quantization) | Intel Neural Compressor, PyTorch |
| 12 | Demo deployment of one of the models developed in the previous chapters | Gradio, PyTorch |
The part of the code was put into the deepdrive_course library to reuse the code and make the notebooks more readable.
All notebooks using deepdrive_course library have the following snippet in the beginning.
import sys
in_colab = "google.colab" in sys.modules
if in_colab:
!git clone https://github.com/abojda/deepdrive_course.git dd_course
!pip install dd_course/ -qIf notebook is run from the Google Colab, this reposity is automatically cloned and library is installed.
To run notebooks locally, deepdrive_course must be installed by hand.
git clone https://github.com/abojda/deepdrive_course.git`pip install -e deepdrive_course/ # Dev installor
pip install deepdrive_course/ # Regular installIt is recommended to use virtual environment (venv, conda, ...) to avoid dependency conflicts with other projects.