.
├── ct
│ ├── data
│ │ - keep each dataset in separate file
│ └── models
│ - keep each model in separate file
│ - concept transformer module will be implemented in some file(s) in this folder?
├── checkpoints
├── logs
│ - tensorboard/wandb logs
├── notebooks
│ - data analysis, model predictions analysis
├── plots
│ - plots for the report
├── trained_models
├── readme.md
└── .gitignore
Download annotations and aYahoo test dataset from: https://vision.cs.uiuc.edu/attributes/ Download aPascal train/validation data from: http://host.robots.ox.ac.uk/pascal/VOC/voc2008/index.html#devkit (Direct dwnld link for aPascal: http://host.robots.ox.ac.uk/pascal/VOC/voc2008/VOCtrainval_14-Jul-2008.tar) VOC2008 documentation downloaded from http://host.robots.ox.ac.uk/pascal/VOC/voc2008/devkit_doc_21-Apr-2008.pdf (We probably do not need it, enough info in attribute_data/README.md file)
https://www.overleaf.com/8966194915vvrcgtbpmpdn
Using Deep Learning VM on google cloud. zone: europe-west1-c series: N1 machine type: n1-highmem-2 (2 vCPU, 13 GB memory) (default) gpu: NVIDIA T4 framework: PyTorch 1.12 (CUDA 11.3) (Don't forget to confirm you want to install GPU drivers) boot disk: Standard Persistent Disk (default) - 200 GB
conda create -n dd2412 python=3.10
conda activate dd2412
conda install pytorch==1.12.1 torchvision==0.13.1 torchaudio==0.12.1 cudatoolkit=11.3 -c pytorch
pip install pytorch_lightning
conda install pandas
<!-- conda install -c conda-forge scikit-learn -->