Visibility Catalog
We classify objects of interest into two categories: Hazards refer to objects that we could trip over or bump into; the visibility of such objects are critical for avoiding safety threatening situations. Reading refers to materials that provide information about the environments, the visibility of which is critical for successful way-finding.
The visibility of any object can change with lighting, materials, size, and the distance and angle the object is being observed. Our dataset includes images and videos of each object with these variations.
Low Vision Ground Truth
Predicting what an individual may or may not see in an environment is the ultimate goal of the EVA project. Visibility ground truth provided by actual low vision individual is critical for our algorithm evaluation in the development stage.
We customize labeling method specifically for low vision individuals to gauge their object localization, identification, and boundary detection performance using their residual vision.
– 2D labeling refers to task performances on static computer images. We introduce tactile interface so that low vision individuals can simply use their fingers to trace out objects.
– 3D labeling refers to task performances in 3D environments. We utilize smart glasses with scene cameras to capture hand gestures while participants attempt to trace out objects in real environments.