June 2020 Summaries

Changelog 6.3 introduces several new features, improvements, and bug fixes across various applications, focusing on enhancing user experience and platform functionality. Notable updates include the asynchronous sending of collaborator emails, consolidation of input-related status codes, and support for adding and training text on the platform. The update also addresses several bugs, such as fixing wrong embed model version IDs in drawing annotations and correcting the swap of predicted and true values in confusion matrix evaluation results. Additionally, enhancements have been made to the Portal, including the creation of a new Single Image View, the ability to display text thumbnails, and improved image tools. The update also introduces a range slider for filtering workflow predictions by value and updates to the Face workflow to include face detection as a concept for search.

Changelog 6.2 introduces a range of new features and improvements across various functionalities, including bug fixes and enhancements specific to enterprise accounts and applications. Key updates include resolving issues with URL errors, application duplication, and input count discrepancies. Significant improvements were made to the task management interface with the creation of a task manager page, new icons, and a unified model structure for consistency. The update also addressed video timeline display issues and improved the demographics model functionality. Additionally, features like apparel detection in demo apps, annotation writing to databases, and concept relation validation have been added, while various deprecated elements have been removed for improved system coherence. Search capabilities have been enhanced, resolving issues with region-specific searches, and the user interface has seen improvements in areas such as collector management and portal tabs. Video prediction workflows have been updated to support better prediction processes, and various internal and public workflows have been patched for enhanced performance.

Clarifai's Release 6.2 introduces a range of innovative tools and enhancements designed to expand AI capabilities within their platform, including advanced model and workflow design tools that facilitate the creation and networking of taxonomies through a Knowledge Graph and concept relations. The new Model Mode allows users to leverage Clarifai's extensive model library or build custom models, offering flexibility in designing AI solutions that require a combination of machine learning and fixed function models. For users prioritizing accuracy and customization, the platform provides prebuilt models with fast training times that can be further adapted through deep training to incorporate new features into workflows. Workflow graphs enable the linking of model outputs to subsequent model inputs, facilitating the development of targeted AI solutions, while the Knowledge Graph offers a structured way to map custom concepts to a shared set of concepts, enhancing interoperability across models through hierarchical relationships based on hypernyms, hyponyms, and synonyms.

Changelog 6.1 introduces a range of new features and improvements, focusing on enhancing platform functionality, refining client-side APIs, and fixing various bugs. Key changes include the removal of Feedback endpoints and the image.crop field across Python, Java, and Javascript clients, as well as the platform and demo portal, simplifying the API structure. The update also adds detection evaluation to the platform and introduces a concept mapping model leveraging knowledge graph relations to enhance user data networks. Bug fixes address performance drops in face predictions and issues with cache invalidation and concept relation deletions. Portal enhancements allow bulk labeling in Explorer mode, display prediction validations, and support multiple concepts per bounding box, while negative tags visibility has been restored. Predict pathway improvements streamline storage interactions and API calls, and video detections now incorporate region predictions from custom models. Additionally, search functionality is expanded with backend support for annotation status searches and connectivity between saved searches and annotation statuses.

Clarifai's Release 6.1 introduces significant enhancements including bug fixes and improvements, notably supporting multi-language search for concepts and enabling the streaming of inputs into the API. The multi-language concept search feature allows users to search and retrieve model predictions in various languages besides English, enhancing accessibility and usability across different linguistic contexts. The streaming inputs capability offers a more efficient way to process visual data by providing an alternative to pagination, allowing users to list and manage application inputs in a scalable and iterative manner, with the process controlled by an optional parameter to specify the starting point of the input list.