Skip to content

Model to check if image was rotated by 90, 180, 270 degrees.

License

Notifications You must be signed in to change notification settings

ternaus/check_orientation

Repository files navigation

Check orientation

Models to check if image was rotated by 0, 90, 180, 270 degrees.

Installation

pip install -U check_orientation

Example inference

Colab notebook with the example: Open In Colab

Training

Define the config.

Example at check_orientation/configs

Define the environmental variable TRAIN_IMAGE_PATH that points to the folder with train dataset.

Example:

export TRAIN_IMAGE_PATH=<path to the tranining folder>

Define the environmental variable VAL_IMAGE_PATH that points to the folder with validation dataset.

Example:

export VAL_IMAGE_PATH=<path to the validation folder>

Training

python -m check_orientation.train -c <path to config>

Inference

python -m torch.distributed.launch --nproc_per_node=<num_gpu> check_orientation/inference.py \
                                   -i <path to images> \
                                   -c <path to config> \
                                   -w <path to weights> \
                                   -o <output-path> \
                                   --fp16

Pre-trained models

Models were pre-trained on the OpenImages dataset.

Models Validation accuracy Config file Weights
swsl_resnext50_32x4d 0.9128 Link Link