Sterblue Curiosity

Creating accurate machine learning models capable of localizing and identifying multiple objects in a single image remains a core challenge in computer vision. Curiosity is an open source framework built on top of Detectron2 that makes it easy to construct, train, evaluate and deploy object detection models.

Why Curiosity ?#

Data versionning#

Data is a huge part of training any neural network, as even the smallest change in your data can have huge consequences on your results. Curiosity allows you to do data versionning and save a json format of your dataset in aws s3, so that your experiments can be reproducible.

Experiment management#

Logging and versionning experiment results is as important as doing the experiments. Curiosity allows for easy integration with aws to log and save your results in an easy human readeable way.

Tensorboard logging#

Tensorboard has become the bread and butter of any DL experiment. Curiosity and thanks to Detectron2 has natif support for tensorboard logging.

Online/Offline training#

Resume your experiments from anywhere, and on any machine. Your results are periodically saved to the cloud, and easily retrievable.

Easy integration#

Integration with Sterblue database for easy recurrent and online training, among other applications.