This thesis paper addresses extensive research on how to augment the traditional data labeling workflow and annotation processing for machine learning and computer vision research, specifically for the development of new AI/ML tools for audio and visual artists.
This paper provides background and history of image classification, applications of supervised learning, and issues with current data labeling approaches for modern artistic applications. There will be an emphasis on the data collection process, the importance of accurate and unbiased data labeling, and the use of open-source software and cloud computing infrastructures for training unique neural networks at scale for knowledge specific tasks.
In order to train comprehensive AI models we need lots of accurately labeled data. This accurately labeled data is often hard to find or difficult to outsource. Label.Art aims to provide artists and researchers with the tools they need to accurately and efficiently label vast amounts of data for machine learning and artificial intelligence computer tasks.
Features of Label.Art include:
Connect your labeling team with a custom or pre-trained ML Backend.
Visualize and compare predictions from different models and perform pre-labeling to speed up the labeling task.
Manage users to collaborate on your data labeling, machine learning, and data science projects.
Cloud Storage: Amazon AWS S3, Google Cloud Storage, or JSON, CSV, TSV, RAR, and ZIP archives.
Custom Annotation Tools