MARLClassification
A PyTorch implementation of a published method for image classification using multi-agent reinforcement learning, where several agents move across an image and their policy is trained to classify it.
The repository reproduces a multi-agent reinforcement-learning approach to image classification, training with an actor-critic method on datasets including MNIST, the NWPU-RESISC45 remote-sensing dataset, the AID aerial-image dataset and WorldStrat satellite imagery. Setup scripts under a resources folder download each dataset, and training and evaluation are run from a single command-line entry point.
Results included in the repository report precision and recall on both the training and evaluation splits for each dataset after 50 training epochs, ranging from about 82% precision and recall on MNIST down to about 56% precision and recall on the harder WorldStrat satellite-imagery task. It requires Python 3.14 or later and the uv package manager.