Taiki Fukiage

Human-like monocular depth biases in deep neural networks

This paper compares human and deep neural network depth judgments, finding that higher accuracy DNNs exhibit human-like biases in depth perception while failing to match human ordinal depth accuracy.

Psychophysical measurement of perceived motion flow of naturalistic scenes

This paper introduces a psychophysical method to measure perceived optical flow in naturalistic scenes and compares it to ground truths and models, revealing flow illusions and computational mechanisms.

Do training with blurred images make convolutional neural networks closer to humans concerning object recognition performance and internal representations?

This study trains CNNs on sharp and blurry images to assess if it improves object recognition robustness and internal representations, finding partial success in reducing texture bias but not achieving human-like performance.