Welcome to my project for CMSC 31230 at the University of Chicago. Please make yourself at home.
-
Clone repository
git clone https://github.com/lawrenceztang/simclr-2 cd simclr-2
-
Install dependencies
conda install --file requirements.txt pip install -U openmim mim install mmcv
-
Download Kinetic Dataset
git clone https://github.com/cvdfoundation/kinetics-dataset.git cd kinetics-dataset bash ./k400_downloader.sh bash ./k400_extractor.sh
-
Run and cancel
python run.py --train_mode=pretrain --train_batch_size=512 --train_epochs=0 --learning_rate=1.0 --weight_decay=1e-4 --temperature=0.5 --dataset=paired --image_size=32 --eval_split=test --resnet_depth=18 --use_blur=False --color_jitter_strength=0.5 --model_dir=/tmp/simclr_test --use_tpu=False --data_dir=paired_dataset
-
Extract image pairs
python3 create_dataset.py
-
Run model
python run.py --train_mode=pretrain --train_batch_size=512 --train_epochs=82 --learning_rate=1.0 --weight_decay=1e-4 --temperature=0.5 --dataset=paired --image_size=32 --eval_split=test --resnet_depth=18 --use_blur=False --color_jitter_strength=0.5 --model_dir=/tmp/simclr_test --use_tpu=False --data_dir=paired_dataset
-
Finetune
python run.py --mode=train_then_eval --train_mode=finetune \ --fine_tune_after_block=4 --zero_init_logits_layer=True \ --variable_schema='(?!global_step|(?:.*/|^)Momentum|head)' \ --global_bn=False --optimizer=momentum --learning_rate=0.1 --weight_decay=0.0 \ --train_epochs=100 --train_batch_size=512 --warmup_epochs=0 \ --dataset=cifar10 --image_size=32 --eval_split=test --resnet_depth=18 \ --checkpoint=/tmp/simclr_test --model_dir=/tmp/simclr_test_ft --use_tpu=False
-
View on Tensorboard
python -m tensorboard.main --logdir=/tmp/simclr_test_ft