Creation of synthetic images, from real images, using a HuggingFace Transformers tool for the subsequent creation of an artificial intelligence model that allows screening between different types of mental illness according to related patterns.
It is simply a non-operational sample demo for demonstration use in the "first AWS Community Builders Hackathon" in June 2023. The API used is simply an example of requesting a certain image but in a real scenario and in production it is they would have to make different requests (one for each disease to be detected) and it should be done with an API capable of doing so) and then distribute the different categories of labeled images to, after synthesizing, train the machine learning model (for example using SageMaker).
The process of creating an artificial intelligence model inexorably depends on the quality of the data. In certain domains, finding an adequate dataset in terms of quantity and quality is not possible. In these cases, generating a dataset by extracting existing data and, later, carrying out a synthesis process can be part of the solution.
The case of mental illnesses and, especially, those that are "detectable" through facial patterns is one of these cases and is absolutely relevant to the case at hand here. In addition, working on artificial intelligence at the service of health, at the service of society, is absolutely motivating.
Through this example project we will be able to check how we can create a good base of original extracted images and, later, synthesize them. This will allow us to create datasets to train artificial intelligence models that would otherwise not be possible for us to build or would require many additional resources to do so.
Main resources:
HuggingFace Transformers
https://huggingface.co/docs/transformers/main/transformers_agents
AWS SageMaker:
https://aws.amazon.com/es/sagemaker/
Github:
https://github.com/gcjordi/HackathonMLAWSCBslack
(Text automatically translated from Spanish to English)
Top comments (0)