Pixano was designed to make the annotation of AI training data for computer vision systems easier. The software recently got some new features that will allow users to do things like explore and analyze datasets and apply data quality processes.
Data-driven AI, a growing trend, focuses on improving the quality of training data first, and optimizing the model second. Pixano software helps with data quality, for better training and, ultimately, more powerful models. The software also supports the development of trustworthy AI.
With Pixano, users can perform statistical calculations on training datasets to determine things like how “noisy” the data is. Data visualization features and filters then make it simple to select which data to include in the training dataset.
Finally, a large language model (LLM) has been integrated into the software’s semantic search engine so that users can retrieve specific images simply by entering a contextual description.
Pixano also offers a range of automatic annotation capabilities that can be applied to images, videos, and point clouds. Third-party algorithms like Meta’s Segment Anything Model (SAM), a deep neural network that can very quickly “cut out” any object from any image, can also be used in Pixano.
Pixano will soon offer model analysis and interpretability features to help users understand and decode the decisions made by the AI.
Pixano can annotate not only images, but also video footage and 2D and 3D point clouds. And it can annotate a mix of different kinds of data. This makes it a very complete software suite that can cover the three main computer vision use cases:
Pixano has been re-released in data scientists’ preferred language, Python, to flatten the learning curve. All of the software’s modules are available under an open source license and can be reused, customized, and added onto. There is also an API that makes Pixano easy to integrate into AI model development workflows and to interface with the most commonly used machine learning frameworks.
Finally, Pixano’s visual components can be brought into Jupyter Notebooks, offering a wide range of customization options.
Pixano gives data scientists the tools they need to ensure data quality through improved annotation. Soon the software will also support more explainable AI models—one of the prerequisites to trustworthy AI. With Pixano, a wide range of use cases will be able to benefit from robust, unbiased, interpretable, documented, quality-data-driven AI.
These capabilities make Pixano a valued tool in CEA-List’s Confiance.ai toolkit. This interoperable platform was set up to facilitate access to the full range of software needed to develop trustworthy AI models. The resources available through Confiance.ai cover all stages of the development cycle, from data acquisition to performance assessment.
Pixano is also compatible with cloud computing standards and integrates seamlessly with MLOps processes.
Pixano is a suite of open-source software modules that we encourage the AI developer community not only to explore, but to expand upon in order to push back the frontiers of the discipline.
