UltraVision

CS 6505 Group Project

installation

This project is set up as a python package to install

conda create --name UltraVision python=3.9
cd path/to/UltraVision
pip install -e .
conda install protobuf

running experiment

Example of running experiment: train simcl

train_bootstrap --label_dir data/FETAL_PLANES_DB_data.csv \
               --data_dir data/Images \
               --data_name FetalPlanes \
               --model_name DenseNet169 \
               --train_strategy train_classification \
               --use_scheduler False \
               --batch_size 32 \
               --train_strategy train_classification \
               --num_epochs  15 \
               --val_size 0.1 \
               --lr 0.00828149 \
               --optimizer_name SGD \
               --one_channel False \
               --save_results_dir UltraVision/experiments/transfer_self_supervised_to_classification \
               --num_bootstraps 5 \
               --freeze_base False

Reproducibility of Results

All experiments discussed in report can be found in the UltraVision/experiments directory. The default seeds in the code are what were used in the experiments.

Tracking Experiments

This project uses both tensorboard and mlflow to track experiments. In order to track experiments with tensorboard, one must set --use_tensorboard to True, mlflow is automatically logged. All training sessions are logged to the working directory.

using tensorboard

tensorboard --logdir path/to/tensorboard/example_20211028_09:32

using mlflow

mlflow ui