October 2020 Summaries

The article provides a detailed guide on using a Python script, developed by a Clarifai engineer, to efficiently upload images and their associated metadata to the Clarifai API. It emphasizes the script's ability to handle data from various sources, performing tasks such as importing region data, managing .csv files with pandas, and utilizing Google protobufs for data translation. The script is designed to batch process data, replacing empty values, parsing metadata, and converting each line into input protos needed for API compatibility. The input_proto format includes image URLs and uses a binary system to denote positive and negative concepts. The article also highlights the use of tools like argparse for command-line argument parsing, ThreadPoolExecutor for multithreading, and tqdm for tracking upload progress. The process culminates in making API requests with authentication metadata, and monitoring response status codes to ensure successful data uploads. For further details, readers are directed to a GitHub repository containing the full script and additional resources.

Clarifai Release 6.9 introduces significant platform enhancements and features designed to optimize AI utilization in business settings. This update includes improvements in data ingestion and labeling, with AI Assist and Auto Annotation enhancements through Scribe, allowing for automatic or human-reviewed data labeling to prevent data drift. The platform has made collectors more user-friendly, enabling automatic data collection every time a model is used, ensuring datasets remain current. Scribe now supports deep training with videos, offering a comprehensive interface for building accurate deep-trained models. Consensus review functionality helps streamline large-scale data annotation by comparing similar annotations for automatic approval, thus enhancing efficiency and quality. The upgraded People and Vehicle Model V2 uses efficientdet d2, providing improved accuracy with fewer resources and is over six times faster than previous models. API improvements include support for secure gRPC channels, which reduce network traffic by approximately 30%, and HTTP/2 support for REST clients, along with a single API domain and better routing management. Additionally, users can now pin important apps on the overview page for easier access and management.