In this paper, we present an end-to-end network, called Cycle-Dehaze, for single image dehazing problem, which does not require pairs of hazy and corresponding ground truth images for training. That is, we train the network by feeding clean and hazy images in an unpaired manner. Moreover, the proposed approach does not rely on estimation of the atmospheric scattering model parameters. Our method enhances CycleGAN formulation by combining cycle-consistency and perceptual losses in order to improve the quality of textural information recovery and generate visually better haze-free images. Typically, deep learning models for dehazing take low resolution images as input and produce low resolution outputs. However, in the NTIRE 2018 challenge on single image dehazing, high resolution images were provided. Therefore, we apply bicubic downscaling. After obtaining low-resolution outputs from the network, we utilize the Laplacian pyramid to upscale the output images to the original resolution. We conduct experiments on NYU-Depth, I-HAZE, and O-HAZE datasets. Extensive experiments demonstrate that the proposed approach improves CycleGAN method both quantitatively and qualitatively.
Cerebrospinal fluid (CSF) flow in the brain ventricles is critical for brain development. Altered CSF flow dynamics have been implicated in congenital hydrocephalus (CH) characterized by the potentially lethal expansion of cerebral ventricles if not treated. CH is the most common neurosurgical indication in children effecting 1 per 1000 infants. Current treatment modalities are limited to antiquated brain surgery techniques, mostly because of our poor understanding of the CH pathophysiology. We lack model systems where the interplay between ependymal cilia, embryonic CSF flow dynamics and brain development can be analyzed in depth. This is in part due to the poor accessibility of the vertebrate ventricular system to
in vivo
investigation. Here, we show that the genetically tractable frog
Xenopus tropicalis
, paired with optical coherence tomography imaging, provides new insights into CSF flow dynamics and role of ciliary dysfunction in hydrocephalus pathogenesis. We can visualize CSF flow within the multi-chambered ventricular system and detect multiple distinct polarized CSF flow fields. Using CRISPR/Cas9 gene editing, we modeled human L1CAM and CRB2 mediated aqueductal stenosis. We propose that our high-throughput platform can prove invaluable for testing candidate human CH genes to understand CH pathophysiology.
Birth defects affect 3% of children in the United States. Among the birth defects, congenital heart disease and craniofacial malformations are major causes of mortality and morbidity. Unfortunately, the genetic mechanisms underlying craniocardiac malformations remain largely uncharacterized. To address this, human genomic studies are identifying sequence variations in patients, resulting in numerous candidate genes. However, the molecular mechanisms of pathogenesis for most candidate genes are unknown. Therefore, there is a need for functional analyses in rapid and efficient animal models of human disease. Here, we coupled the frog Xenopus tropicalis with Optical Coherence Tomography (OCT) to create a fast and efficient system for testing craniocardiac candidate genes. OCT can image cross-sections of microscopic structures in vivo at resolutions approaching histology. Here, we identify optimal OCT imaging planes to visualize and quantitate Xenopus heart and facial structures establishing normative data. Next we evaluate known human congenital heart diseases: cardiomyopathy and heterotaxy. Finally, we examine craniofacial defects by a known human teratogen, cyclopamine. We recapitulate human phenotypes readily and quantify the functional and structural defects. Using this approach, we can quickly test human craniocardiac candidate genes for phenocopy as a critical first step towards understanding disease mechanisms of the candidate genes.
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.