Mapika released a compact vision model decider-2b-vision

Developers have introduced decider-2b-vision, a specialized neural network for computer vision tasks, designed to classify images with high accuracy. The model weighs 4.1 GB and is intended for tasks that require choosing from several answer options.

The model’s architecture is optimized to complete tasks in a single forward pass. During testing, the system achieved an accuracy of 0.89 on the Visual7W dataset. When analyzing game frames in Pong and Breakout, its accuracy reached 0.96.

A key feature of the model is its ability to provide calibrated probabilities for each proposed answer option, identified by a letter. This makes it suitable for scenarios that require not just object recognition but decision-making based on visual data.

The model is available for download on Hugging Face. Its compact size allows it to run on consumer hardware, simplifying integration into local image analysis systems and automated classification tools.