digital-nomad-cheng/iris_landmarks_pytorch

Iris landmarks localization.

Dense Iris Landmarks

Repo for dense iris landmarks localization with synthesized eye dataset.

Repo Structure

./config.py: config file

./loss.py: loss function

./checkpoint.py: save the trained model

./tools: some utitilies

./test: test results

Prepare Dataset

There are two methods to prepare the training data.

1. You could use the software here to synthesiz all kinds of data yourself. Then use scripts in ./gen_dataset to generate training data.

2. You could also use the dataset I provided. directly. Just download the dataset and put train images in ./data directory. In this case the annotations are already prepared in annotations directory.

   Adress:https://pan.baidu.com/s/1gzYAVvEuhuu6L8tos3zXAQ
   Password:990n

Train

config all the training parameters in config.py RUN python training/train.py to train your model.

Training results are kept in results directory.

Test

Put well croped eye images in ./data/test RUN python test/test_image.py to test your model.

iris_landmarks_pytorch's Issues

The training loss suddenly increases at the beginning of each round

Data issue

Hello, I'm an independent researcher trying to replicate your model for research purpose.
I cannot get access to your dataset, which requires Baidu account.
Baidu account recently requires Chinese phone number.

May I get access to your dataset in other way?

thanks.