SyncMapV2 is an unsupervised learning-based image segmentation algorithm that shows much better robustness than existing state-of-the-art (SOTA) algorithms. Even in images with digital damage (noise, weather effects, blur), the mean intersection over union (mIoU) degradation is very small (0.01%), which is significantly better than the degradation rate of SOTA algorithms (23.8%). It is based on a learning paradigm that combines self-organizing dynamics equations and random network concepts without robust training, supervision, or loss function. In addition, unlike existing methods, it adapts online without the need for reinitialization for each input, mimicking the continuous adaptability of human vision. Thus, beyond accurate and robust results, we present the first algorithm that adapts online. Adaptability tests also show almost no performance degradation.
