Abstract
The CarML (Cognitive ARtifacts for Machine Learning) is a platform allowing people to easily deploy and experiment ML (Machine Learning)/DL (Deep Learning) frameworks and models. It allows ML/DL software developers to deploy their software packages, ML model developers to publish and evaluate their models, users to experiment with different models and frameworks, all through a web user interface or a REST api, and system architects to capture system resource usage to inform future system and hardware configuration. In order to compare the accuracy that a model predicts among frameworks, instead of using a existing model from different framework, it is better to use a model in one framework to convert to other frameworks. There are many existing converters for framework conversion; however, none of them is able to convert all kinds of models.
Therefore, I researched on how to solve the issues happened when converting from Caffe to TensorFlow, and composed a script to automatically collect and analyze results by inquiring CarML by images.
