resnet-18-tensorflow

A TensorFlow implementation of ResNet-18(https://arxiv.org/abs/1512.03385)

Prerequisite

1. TensorFlow 1.8

2. The ImageNet dataset

• All image files are required to be valid JPEG files. See this gist.

• It is highly recommened for every image to be resized so that the shorter side is 256.

3. (Optional) Torchfile(to convert ResNet-18 .t7 checkpoint into tensorflow checkpoint. Install with a command pip install torchfile)

How To Run

• (Optional) Convert torch .t7 into tensorflow ckpt

# Download the ResNet-18 torch checkpoint
wget https://d2j0dndfm35trm.cloudfront.net/resnet-18.t7
# Convert into tensorflow checkpoint
python extract_torch_t7.py

1. Modify train_scratch.sh(training from scratch) or train.sh(finetune pretrained weights) to have valid values of following arguments

• train_dataset, train_image_root, val_dataset, val_image_root: Path to the list file of train/val dataset and to the root

• num_gpus and corresponding IDs of GPUs(CUDA_VISIBLE_DEVICES at the first line)

2. Run!

• ./train.sh if you want to finetune the converted ResNet(NOTE: The model needs to be finetuned for some epochs)

• ./train_scratch.sh if you want to train ResNet from scratch

3. Evaluate the trained model

• ./eval.sh for evaluating the trained model(change the arguments in eval.sh to your preference)

Note

• The extracted weights should be finetuned for several epochs(run ./train.sh) to get the full performance(If you run the evaluation code without finetuning, the single-crop top-1 validation accuracy is about 60%, which is less than the appeared in the original). I guess there is some minor issue that I have missed.