- Project Zeta
Using a modular approach, this algorithm supports true multi-modal pipelining where different modes of data can be independently processed and combined to produce a knowledge-rich, intercorrelated database that can abstractify complex situational data. This allows for constant expansion of new sensors and data streams that readily integrate with existing ones allowing for a dynamic, ever-evolving data structure. - Unsupervised Visual Segmentation
Using a k-means clustering approach, this algorithm intelligently identifies key components of the image that can be processed later or reverse-searched to find more information. Using cluster refinement, the algorithm can identify the most important features of the image and remove extraneous segments. After reverse-searching these images, the image category is extracted and the Google Images API is utilized to download hundreds more images of the same class to train various models. - Fast Label-Less Object Detection
Using a novel approach to object detection, this algorithm can identify objects in images without the need for labeled data. By decomposing the image and using image recognition analysis to segments compared to the whole image, MAGIST can approximate image bounding boxes. This abstractifies the data into a more condensed and knowledge rich form that is easily processable or stored in the memory models. - Project ZetaUsing a modular approach, this algorithm supports true multi-modal pipelining where different modes of data can be independently processed and combined to produce a knowledge-rich, intercorrelated database that can abstractify complex situational data. This allows for constant expansion of new sensors and data streams that readily integrate with existing ones allowing for a dynamic, ever-evolving data structure.
- Unsupervised Visual SegmentationUsing a k-means clustering approach, this algorithm intelligently identifies key components of the image that can be processed later or reverse-searched to find more information. Using cluster refinement, the algorithm can identify the most important features of the image and remove extraneous segments. After reverse-searching these images, the image category is extracted and the Google Images API is utilized to download hundreds more images of the same class to train various models.
1 of 3
Development
This project released its first major version v0.1.0 AlphaBird on June 26, 2022 with current development leading to integration of this algorithm with the RaSDAN Algorithm to finish a multi-modal hippocampal model. With source code and Python wheels released, the software is highly accessible for any developer to contribute to.
Features
In its current state, MAGIST can use chronological visual information and audio information and process them individually. For visual information, the algorithm can segment objects in the foreground fully unsupervised and used this information to scrape search engines for a potential label. This can be used to train a unsupervised, custom object detector while also scraping the internet for more information regarding the label. The full audio pipeline is still under development but just uses a speech-to-text engine to correlate speech with the visual stimuli.