Treasure Hunter performs grading, attribution and authentication of collectibles, such as trading and baseball cards, coins and comic books, objectively and at scale with human-level accuracy. Treasure Hunter boosts throughput that translates directly to revenue growth and introduces novel subscription-based profit centers that were previously unimaginable.
Fine art, jewelry, watches, and wine much like the financial market go up and down in price. By measuring; how the market values an item, the proclivity of the buyers, economic conditions, and other indicative features you can accurately assess the desirability, demand, and thus value of an item. Furthermore, you can quantify the authentic engagement of the market to surface predictive insights into the value of an item.
Replicates with human accuracy the work of expert human graders dramatically reducing repetitive work and error rates while simultaneously increasing overall throughput.
Replenish and augment a sparse and difficult to train talent pool at expert level accuracy.
Detailed reasoning of attribution and explainability of AI generated grading scores provides the transparency required to increase consumer confidence and engagement with outputs.
Extract the collectible piece from an image and passes only the object to the grading model, ensuring no unnecessary artifacts are introduced into the grading process.
Using an ensemble method we train a series of deep learning models to generate low dimensional representations of all images in the database storing both local and global features. Low level representations of new images can then be compared to this database and efficiently attributed to the correct object using a nearest neighbor search.
Image centering, corner quality, edge quality, and surface texture all generate independent scores using a multi-task learning algorithm.
Merging human defined feature scores together with the region of interest extraction in a residual neural network model, the system outputs a grading score for each object on par with expert human level graders.
As the continuously learning visual AI ingests additional images, the accuracy of the system continues improving. Accrete’s continuous learning technology has the unique ability to adapt dynamically, like a human, learning and becoming smarter as it incorporates new information and contexts.
One continuous learning model improves accuracy and handles various image qualities, additional artifacts, and object types through transfer learning.