Object recognition and computer vision 2021/2022

Assignment 3: Image classification

Create virtual environment and install requirements

python3 -m venv env
source env/bin/activate

pip install torch # necessary to install a first version of it because of Detectron2
pip install -r requirements.txt

Dataset

We will be using a dataset containing 200 different classes of birds adapted from the CUB-200-2011 dataset. Download the training/validation/test images from here. The test image labels are not provided.

Small EDA on the data

Take a look at the notebook EDA.ipynb.

Report

The overall method and results are described in the report.pdf.

Run Mask R-CNN on all images and save newly cropped data

python3 -m main_detector.py

Run feature extractor to get 2048-feature vector for each image

python3 -m main_feature_extractor.py

Examples on how to use these embeddings can be found in the notebooks demo.

Train the selected model on train set and evaluate it on val set

This script will train the defined model without previously computed embeddings.

python3 -m main_classifier_without_embeddings.py

Train the selected model using Cross Validation

Using cross validation (CV) to train the selected model.

python3 -m main_kfolds.py

Acknowledgments

Adapted from Rob Fergus and Soumith Chintala https://github.com/soumith/traffic-sign-detection-homework.
Adaptation done by Gul Varol: https://github.com/gulvarol