Predicting Perceived Gloss: Do Weak Labels Suffice?
This work addresses the costly and limited generalization of human annotations for material perception in computer vision, offering an incremental improvement in efficiency.
The paper tackles the problem of predicting human gloss perception from images by augmenting a small set of human-annotated strong labels with automatically derived weak labels, resulting in enhanced gloss prediction beyond state-of-the-art and a substantial reduction in annotation costs without sacrificing accuracy.
Estimating perceptual attributes of materials directly from images is a challenging task due to their complex, not fully-understood interactions with external factors, such as geometry and lighting. Supervised deep learning models have recently been shown to outperform traditional approaches, but rely on large datasets of human-annotated images for accurate perception predictions. Obtaining reliable annotations is a costly endeavor, aggravated by the limited ability of these models to generalise to different aspects of appearance. In this work, we show how a much smaller set of human annotations ("strong labels") can be effectively augmented with automatically derived "weak labels" in the context of learning a low-dimensional image-computable gloss metric. We evaluate three alternative weak labels for predicting human gloss perception from limited annotated data. Incorporating weak labels enhances our gloss prediction beyond the current state of the art. Moreover, it enables a substantial reduction in human annotation costs without sacrificing accuracy, whether working with rendered images or real photographs.