From Images to Perception: Emergence of Perceptual Properties by Reconstructing Images
By: Pablo Hernández-Cámara, Jesus Malo, Valero Laparra
Potential Business Impact:
Computer sees images like humans do.
A number of scientists suggested that human visual perception may emerge from image statistics, shaping efficient neural representations in early vision. In this work, a bio-inspired architecture that can accommodate several known facts in the retina-V1 cortex, the PerceptNet, has been end-to-end optimized for different tasks related to image reconstruction: autoencoding, denoising, deblurring, and sparsity regularization. Our results show that the encoder stage (V1-like layer) consistently exhibits the highest correlation with human perceptual judgments on image distortion despite not using perceptual information in the initialization or training. This alignment exhibits an optimum for moderate noise, blur and sparsity. These findings suggest that the visual system may be tuned to remove those particular levels of distortion with that level of sparsity and that biologically inspired models can learn perceptual metrics without human supervision.
Similar Papers
Leveraging Geometric Visual Illusions as Perceptual Inductive Biases for Vision Models
CV and Pattern Recognition
Teaches computers to see better using optical illusions.
Visual Image Reconstruction from Brain Activity via Latent Representation
CV and Pattern Recognition
Shows what people see from brain signals.
Improving VisNet for Object Recognition
CV and Pattern Recognition
Helps computers see and recognize objects like humans.