AtheMathmo / resnet

Modified from the original tensorflow version.

resnet

Standard ResNet training on image classification benchmarks. Modified from the original tensorflow version.

Installation

Custom paths first in setup.sh (data folder, model save folder, etc.).

git clone --recursive git://github.com/renmengye/resnet.git
cd resnet
./setup.sh

CIFAR-10/100

./run_cifar_exp.py --dataset cifar-10 --model resnet-32

ImageNet

# Run training.
./run_imagenet_exp.py --model resnet-50

# Evaluate a trained model. Launch this on a separate GPU. 
./run_imagenet_eval.py --id [EXPERIMENT ID]

ImageNet on Slurm

SSH into the slurm manager node first, and then launch jobs there.

# Launch a recurring training job, 30K steps per job, for total 600K steps.
./run_imagenet_exp_sched.py --model resnet-50 --max_num_steps 30000 --max_max_steps 600000

# Launch a recurring evaluation job every 2 hours.
./run_imagenet_eval_sched.py --id [EXPERIMENT ID] --min_interval 7200

Provided Model Configs

See resnet/configs/cifar_exp_config.py and resnet/configs/imagenet_exp_config.py