A Benchmark for Burst Color Constancy

Abstract

Temporal Color Constancy (CC) is a recently proposed approach that challenges the conventional single-frame color constancy. The conventional approach is to use a single frame -shot frame -to estimate the scene illumination color. In temporal CC, multiple frames from the view finder sequence are used to estimate the color. However, there are no realistic large scale temporal color constancy datasets for method evaluation. In this work, a new temporal CC benchmark is introduced. The benchmark comprises of (1) 600 real-world sequences recorded with a high-resolution mobile phone camera, (2) a fixed train-test split which ensures consistent evaluation, and (3) a baseline method which achieves high accuracy in the new benchmark and the dataset used in previous works. Results for more than 20 well-known color constancy methods including the recent state-of-the-arts are reported in our experiments.

Cascading Convolutional Temporal Colour Constancy

“…As part of this analysis, we investigate the number of frames involved in the illuminant estimation as a possible way to reduce the inference time of the TCC models. Contrary to preliminary results in [19], we show that it is possible to train models that work with shorter sequences, with substantial gain in inference efficiency and no major loss in accuracy, if the retained frames are selected to capture global temporal information.…”

“…Most of the research so far has focused on CCC for single frames (see [11] for a survey), with some attempts of leveraging the temporal information intrinsic in sequences of correlated frames for illuminant estimation [18,19]. In both cases, state-of-the-art methods in terms of accuracy rely on Deep Learning (DL).…”

“…In this paper, we investigate how to advance the stateof-the-art in TCC by proposing three new architectures that integrate the current best-performing model for TCC [19] with a strategy that proved successful in the single-frame scenario [29] but which is novel to the temporal task. Our results show that the novel approaches surpass in terms of accuracy the state-of-the-art on the TCC benchmark [19], which is currently the only dataset specifically designed for experimenting with CCC in the temporal scenario. We also present a detailed analysis of the trade-off between model accuracy and computational cost (in terms of inference time and memory occupancy), relevant for the practical applications of CCC.…”

Point Cloud Color Constancy

“…Barron et al [9] proposed a Fourier transform-based model (model O in the paper), learning the model weight from the camera metadata like aperture and so on. Qian et al [33,35] claimed the preceding image sequence benefits the illumination estimation for the shot image. Yoo et al [45] explored the AC light source in a high-speed setting.…”

SDE-AWB: a Generic Solution for 2nd International Illumination Estimation Challenge

“…Bianco et al [10] initially shows the potential of applying CNN for illumination estimation. Then more effective and efficient CNN variants are designed, like AlexNet+SVM [11], DS-Net [12], FC4-Net [13], RCC [14], BCC [15] and so on.…”