We present a new approach to the video matting problem which improves the temporal coherency. In existing video matting approaches the alpha matte sequence is determined frame by frame, which leads to flickering near the boundary of the extracted foreground region. Especially the temporal artifacts arise when an alpha matte has some errors. We reduce this effect by considering several consecutive frames in obtaining the alpha matte. Results show improved temporal coherence and an accurate alpha matte.
The prediction reliability of neural networks is important in many applications. Specifically, in safety-critical domains, such as cancer prediction or autonomous driving, a reliable confidence of models prediction is critical for the interpretation of the results. Modern deep neural networks have achieved a significant improvement in performance for many different image classification tasks. However, these networks tend to be poorly calibrated in terms of output confidence. Temperature scaling is an efficient postprocessing-based calibration scheme and obtains well calibrated results. In this study, we leverage the concept of temperature scaling to build a sophisticated bin-wise scaling. Furthermore, we adopt augmentation of validation samples for elaborated scaling. The proposed methods consistently improve calibration performance with various datasets and deep convolutional neural network models.
We introduce a stable noise function with controllable properties. The well-known Perlin noise function is generated by interpolation of a pre-defined random number table. This table must be modified if user-defined constraints are to be satisfied, but modification can destroy the stability of the table. We integrate statistical tools for measuring the stability of a random number table with user constraints within an optimization procedure, so as to create a controlled random number table which nevertheless has a uniform random distribution, no periodicity, and a band-limited property.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.