Learning visual biases from human imagination
This work addresses the challenge of improving object recognition in computer vision, particularly in low-data scenarios, though it appears incremental in its application of existing psychophysics tools.
The paper tackles the problem of extracting biases from the human visual system and transferring them to machine recognition systems, showing that this approach can help object recognition generalize across datasets and perform better with limited training data.
Although the human visual system can recognize many concepts under challenging conditions, it still has some biases. In this paper, we investigate whether we can extract these biases and transfer them into a machine recognition system. We introduce a novel method that, inspired by well-known tools in human psychophysics, estimates the biases that the human visual system might use for recognition, but in computer vision feature spaces. Our experiments are surprising, and suggest that classifiers from the human visual system can be transferred into a machine with some success. Since these classifiers seem to capture favorable biases in the human visual system, we further present an SVM formulation that constrains the orientation of the SVM hyperplane to agree with the bias from human visual system. Our results suggest that transferring this human bias into machines may help object recognition systems generalize across datasets and perform better when very little training data is available.