ShapeNoiseHorseBird

The ShapeNoiseHorseBird dataset is a curated collection designed to challenge shape recognition models with varying levels of noise. It integrates samples from the Weizmann Horse and Caltech-UCSD Birds 200 datasets, diversifying the challenges presented to models. The introduced noise contains Salt and pepper noise, Circle Noise, Real image Noise, Occlusion noise, Thresholded probability noise and Detection image Noise, creating a robust benchmark for evaluating the resilience of shape-based recognition algorithms. Researchers can leverage ShapeNoise to assess the adaptability of their models to scenarios where shape perception is crucial.