Limits of peripheral acuity and implications for VR system design

Abstract

At different retinal locations, we measured the visual system's sensitivity to a number of artifacts that can be introduced in near‐eye display systems. One study examined the threshold level of downsampling that an image can sustain at different positions in the retina and found that temporally stable approaches, both blurred and aliased, were much less noticeable than temporally volatile approaches. Also, boundaries between zones of different resolution had low visibility in the periphery. We also examined the minimum duration needed for the visual system to detect a low‐resolution region in an actively tracked system and found that low‐resolution images presented for less than 40 ms before being replaced with a high‐resolution image are unlikely to be visibly degraded. We also found that the visual system shows a rapid falloff in its ability to detect chromatic aberration in the periphery. These findings can inform the design on high performance and computationally efficient near‐eye display systems.

Luminance-contrast-aware foveated rendering

“…Our technique relies on two critical observations described below. Hoffman et al [2018] and Albert et al [2017] demonstrate that temporarily-stable low-resolution rendering is perceptually equivalent to a Gaussian-blurred high-resolution rendering. This motivates our technique to model the resolution reduction using a Gaussian low-pass filter.…”

“…Both approaches employ an implementation of our predictor in the underlying shader language, which guides the selection of spatially varying rendering bandwidth. The use of Gaussian filter for simulated foveation in the Unity3D implementation provides a good temporal stability, and it was demonstrated to be visually similar to an anti-aliasing which avoids spurious artifacts resulting from undersampling [Albert et al 2017;Hoffman et al 2018]. By using this strategy, we make sure that the decisions of participants are primarily based on the visual evaluation of foveation quality throughout our experiments and isolated from other external factors such as the quality of a particular temporal antialiasing routine that might be employed in the pipeline.…”

Toward Standardized Classification of Foveated Displays

“…Interestingly, Hoffman et al reported that in the non-foveated, acuity matched situation (class 4A), a 2-tiered foveated display is less noticable than a N-tiered foveated display if the transition occurs in the periphery region [7]. Meanwhile, in practically foveated (Class 2) designs with gaze tracking [15,25], N-tiered display provided the same level of perceptual quality to the uniform resolution display.…”

Aligning content rendering resolution and feature size with display capability in near‐eye display systems

“…This is consistent with other studies looking at text rendering in VR headsets that found size is limited by pixel arrangement . Temporal volatility that is typical of near‐eye display systems can enhance otherwise minimally visible pixel level changes . Selection of appropriate fonts and rendering strategy is widely suggested as an important development decision, but guidance on some best practices can differ depending on the headsets .…”

“…11 Temporal volatility that is typical of near-eye display systems can enhance otherwise minimally visible pixel level changes. 12,13 Selection of appropriate fonts and rendering strategy is widely suggested as an important development decision, but guidance on some best practices can differ depending on the headsets. 14,15 For example, FIGURE 1 Importance of appropriate text sizing and layout for good experience on tier 1 (high-density) and tier 2 (mid-density) displays most text-based guidelines assume the text will be well aligned with the pixel grid; however, in a near-eye headset, world-tracked text is aligned to a spatial position in space and is resampled to the pixel array.…”