MatMorph Material Transformation dataset

MatMorph is and Material Transformation benchmark comprising of image sequences from a wide rangeof real-world processes and transformations, including cleaning, folding, cooking, rusting, cracking, melting, burning, sprouting, growing, mixing, dissolving, bleaching, swelling, and many others. These processes were recorded in both indoor and outdoor environments. For each transformation sequence, we tracked one object, for example, drying leaves, rusting metals or melting ice. In a typical procedure, we allowed a change to become clearly visible, then moved the object toa new indoor or outdoor location and photographed it from a different angle. The process was then allowed to continue until the next noticeable change occurred, and the cycle was repeated. This resulted in a sequences of images that captures a gradual transformation of the same object, with each frame taken under different positional, environmental, and lighting conditions. For immobile systems, such as walls, floors, or other fixed surfaces, where physical relocation is not possible, we created a separate subset termed the ”immobile setting.” This subset is used as a distinct test set, and even in these cases, we varied the camera position to maximize visual variability.For each image there is associated mask image of the region of the object/material in which the transformation occur.