September 2020 Summaries

Clarifai's new Data Mode in the Clarifai Portal simplifies the process of loading data into the platform for labeling, training, searching, and predicting, offering features for uploading images, videos, and text either from URLs or directly from a computer. It supports CSV and TSV file uploads, making it convenient for managing multiple inputs, but when working with extremely large datasets or integrating data ingestion within an application pipeline, the Clarifai API becomes essential. The API can be configured using a Python gRPC client to upload images and associated data, with a provided script illustrating the process, including parsing CSV files into data frames, managing metadata, and handling inputs through multithreading and batch processing. The implementation involves using various Python libraries to manage data and API requests efficiently, with detailed steps for constructing and processing input data protos, making API calls, and handling potential errors, while ensuring seamless integration into existing workflows.

Clarifai Release 6.8 introduces new features designed to enhance AI-powered business solutions, including Data Mode, which simplifies data loading for AI projects, and state-of-the-art people and vehicle detectors for security and analysis applications. The release also features a rebuilt general detector based on the Open Images Dataset, offering enriched annotations and complex scene analysis. A new text aggregation operator enhances visual text recognition by automating the arrangement of words into sentences. Additionally, the update includes the addition of C# and PHP to the client library, catering to popular programming languages used in diverse domains such as game development and server-side scripting.

The text discusses a new approach to visual recognition of race, age, and gender, designed to minimize racial bias in AI models. By separating these categories into individual models and then integrating them into a comprehensive Demographics Workflow, the system offers enhanced flexibility and easier data interpretation. The previous models were often biased towards Caucasian faces due to skewed training data, which affected their accuracy across various demographic groups. The new model is built on a diverse dataset representing seven racial groups, ensuring balanced representation and consistent accuracy across races and genders. This Demographics Workflow introduces changes in API calls and response formats, providing more reliable demographic insights.