Browsing by Author "Fux, Michal"

Now showing 1 - 2 of 2

Comparing Humans and Deep Neural Networks on Face Recognition Under Various Distance and Rotation Viewing Conditions
(Journal of Vision, 2023) Fux, Michal; Arslan , Şuayb Şefik; Jang, Hojin; Boix, Xavier; Cooper, Avi; Groth, Matt J; Sinha, Pawan
Humans possess impressive skills for recognizing faces even when the viewing conditions are challenging, such as long ranges, non-frontal regard, variable lighting, and atmospheric turbulence. We sought to characterize the effects of such viewing conditions on the face recognition performance of humans, and compared the results to those of DNNs. In an online verification task study, we used a 100 identity face database, with images captured at five different distances (2m, 5m, 300m, 650m and 1000m) three pitch values (00 - straight ahead, +/- 30 degrees) and three levels of yaw (00, 45, and 90 degrees). Participants were presented with 175 trials (5 distances x 7 yaw and pitch combinations, with 5 repetitions). Each trial included a query image, from a certain combination of range x yaw x pitch, and five options, all frontal short range (2m) faces. One was of the same identity as the query, and the rest were the most similar identities, chosen according to a DNN-derived similarity matrix. Participants ranked the top three most similar target images to the query image. The collected data reveal the functional relationship between human performance and multiple viewing parameters. Nine state-of-the-art pre-trained DNNs were tested for their face recognition performance on precisely the same stimulus set. Strikingly, DNN performance was significantly diminished by variations in ranges and rotated viewpoints. Even the best-performing network reported below 65% accuracy at the closest distance with a profile view of faces, with results dropping to near chance for longer ranges. The confusion matrices of DNNs were generally consistent across the networks, indicating systematic errors induced by viewing parameters. Taken together, these data not only help characterize human performance as a function of key ecologically important viewing parameters, but also enable a direct comparison of humans and DNNs in this parameter regime
What Is the Effective Resolution of the Retinal Image of a Distant Face?
(Vision Sciences Society Annual Meeting Abstract, 2023) Arslan , Şuayb Şefik; Fux, Michal; Sinha, Pawan
We consider the following question: What is the effective resolution of a face image projected on the retina, when the face is at a specified distance from the eye? Though simple to state, this is a surprisingly challenging issue to resolve. The mapping between viewing distance and effective resolution cannot be readily derived based on the contrast sensitivity, Snellen acuity, or even the packing density of photoreceptors in the fovea. With initial guidelines derived from theoretical considerations, images of varying resolution were presented across a range of viewing distances. For each distance, participants were required to perform an ‘odd one out’ task. This involved detecting the one that was different from the rest in a 2x2 grid, with image resolution being the only dimension of variation. As the experiment progressed, the viewing distance decreased monotonically, and participants were able to detect increasingly subtle resolution differences between the three standard images and the outlier. The collected data have allowed us to establish the upper/lower bounds on the effective available resolution for typical human vision as a function of viewing distance. Interestingly, we find that humans perform significantly better, particularly at short ranges, than what a theoretical model predicts based on projected image size, cone density, and foveal extent. Accordingly, we suggest that the non-uniform in-fovea density, as well as less sharp fall-off in the acuity density function outside the fovea, need to be integrated into future theoretical models to translate viewing distance to perceived image characteristics. A pragmatic benefit of the mapping is that it enables a direct comparison of human face recognition performance as assessed across blur and viewing distance. Additionally, it allows us to systematically compare human performance on face recognition at varying distances with that of machine vision systems using the common axis of resolution.