Bridging the Gap Between Computational Photography and Visual Recognition.

What is the current state-of-the-art for image restoration and enhancement applied to degraded images acquired under less than ideal circumstances? Can the application of such algorithms as a pre-processing step improve image interpretability for manual analysis or automatic visual recognition to classify scene content? While there have been important advances in the area of computational photography to restore or enhance the visual quality of an image, the capabilities of such techniques have not always translated in a useful way to visual recognition tasks. Consequently, there is a pressing need for the development of algorithms that are designed for the joint problem of improving visual appearance and recognition, which will be an enabling factor for the deployment of visual recognition tools in many real-world scenarios. To address this, we introduce the UG 2 dataset as a large-scale benchmark composed of video imagery captured under challenging conditions, and two enhancement tasks designed to test algorithmic impact on visual quality and automatic object recognition. Furthermore, we propose a set of metrics to evaluate the joint improvement of such tasks as well as individual algorithmic advances, including a novel psychophysics-based evaluation regime for human assessment and a realistic set of quantitative measures for object recognition performance. We introduce six new algorithms for image restoration or enhancement, which were created as part of the IARPA sponsored UG 2 Challenge workshop held at CVPR 2018. Under the proposed evaluation regime, we present an in-depth analysis of these algorithms and a host of deep learning-based and classic baseline approaches. From the observed results, it is evident that we are in the early days of building a bridge between computational photography and visual recognition, leaving many opportunities for innovation in this area.

Clustering Millions of Faces by Identity.

Given a large collection of unlabeled face images, we address the problem of clustering faces into an unknown number of identities. This problem is of interest in social media, law enforcement, and other applications, where the number of faces can be of the order of hundreds of million, while the number of identities (clusters) can range from a few thousand to millions. To address the challenges of run-time complexity and cluster quality, we present an approximate Rank-Order clustering algorithm that performs better than popular clustering algorithms (k-Means and Spectral). Our experiments include clustering up to 123 million face images into over 10 million clusters. Clustering results are analyzed in terms of external (known face labels) and internal (unknown face labels) quality measures, and run-time. Our algorithm achieves an F-measure of 0.87 on the LFW benchmark (13 K faces of 5,749 individuals), which drops to 0.27 on the largest dataset considered (13 K faces in LFW + 123M distractor images). Additionally, we show that frames in the YouTube benchmark can be clustered with an F-measure of 0.71. An internal per-cluster quality measure is developed to rank individual clusters for manual exploration of high quality clusters that are compact and isolated.

Face Search at Scale.

Given the prevalence of social media websites, one challenge facing computer vision researchers is to devise methods to search for persons of interest among the billions of shared photos on these websites. Despite significant progress in face recognition, searching a large collection of unconstrained face images remains a difficult problem. To address this challenge, we propose a face search system which combines a fast search procedure, coupled with a state-of-the-art commercial off the shelf (COTS) matcher, in a cascaded framework. Given a probe face, we first filter the large gallery of photos to find the top- k most similar faces using features learned by a convolutional neural network. The k retrieved candidates are re-ranked by combining similarities based on deep features and those output by the COTS matcher. We evaluate the proposed face search system on a gallery containing 80 million web-downloaded face images. Experimental results demonstrate that while the deep features perform worse than the COTS matcher on a mugshot dataset (93.7 percent versus 98.6 percent TAR@FAR of 0.01 percent), fusing the deep features with the COTS matcher improves the overall performance ( 99.5 percent TAR@FAR of 0.01 percent). This shows that the learned deep features provide complementary information over representations used in state-of-the-art face matchers. On the unconstrained face image benchmarks, the performance of the learned deep features is competitive with reported accuracies. LFW database: 98.20 percent accuracy under the standard protocol and 88.03 percent TAR@FAR of 0.1 percent under the BLUFR protocol; IJB-A benchmark: 51.0 percent TAR@FAR of 0.1 percent (verification), rank 1 retrieval of 82.2 percent (closed-set search), 61.5 percent FNIR@FAR of 1 percent (open-set search). The proposed face search system offers an excellent trade-off between accuracy and scalability on galleries with millions of images. Additionally, in a face search experiment involving photos of the Tsarnaev brothers, convicted of the Boston Marathon bombing, the proposed cascade face search system could find the younger brother's (Dzhokhar Tsarnaev) photo at rank 1 in 1 second on a 5 M gallery and at rank 8 in 7 seconds on an 80 M gallery.

Demographic Estimation from Face Images: Human vs. Machine Performance.

Demographic estimation entails automatic estimation of age, gender and race of a person from his face image, which has many potential applications ranging from forensics to social media. Automatic demographic estimation, particularly age estimation, remains a challenging problem because persons belonging to the same demographic group can be vastly different in their facial appearances due to intrinsic and extrinsic factors. In this paper, we present a generic framework for automatic demographic (age, gender and race) estimation. Given a face image, we first extract demographic informative features via a boosting algorithm, and then employ a hierarchical approach consisting of between-group classification, and within-group regression. Quality assessment is also developed to identify low-quality face images that are difficult to obtain reliable demographic estimates. Experimental results on a diverse set of face image databases, FG-NET (1K images), FERET (3K images), MORPH II (75K images), PCSO (100K images), and a subset of LFW (4K images), show that the proposed approach has superior performance compared to the state of the art. Finally, we use crowdsourcing to study the human perception ability of estimating demographics from face images. A side-by-side comparison of the demographic estimates from crowdsourced data and the proposed algorithm provides a number of insights into this challenging problem.

The time dependent association of adrenaline administration and survival from out-of-hospital cardiac arrest.

BACKGROUND: Recommended for decades, the therapeutic value of adrenaline (epinephrine) in the resuscitation of patients with out-of-hospital cardiac arrest (OHCA) is controversial. PURPOSE: To investigate the possible time-dependent outcomes associated with adrenaline administration by Emergency Medical Services personnel (EMS). METHODS: A retrospective analysis of prospectively collected data from a near statewide cardiac resuscitation database between 1 January 2005 and 30 November 2013. Multivariable logistic regression was used to analyze the effect of the time interval between EMS dispatch and the initial dose of adrenaline on survival. The primary endpoints were survival to hospital discharge and favourable neurologic outcome. RESULTS: Data from 3469 patients with witnessed OHCA were analyzed. Their mean age was 66.3 years and 69% were male. An initially shockable rhythm was present in 41.8% of patients. Based on a multivariable logistic regression model with initial adrenaline administration time interval (AATI) from EMS dispatch as the covariate, survival was greatest when adrenaline was administered very early but decreased rapidly with increasing (AATI); odds ratio 0.94 (95% Confidence Interval (CI) 0.92-0.97). The AATI had no significant effect on good neurological outcome (OR=0.96, 95% CI=0.90-1.02). CONCLUSIONS: In patients with OHCA, survival to hospital discharge was greater in those treated early with adrenaline by EMS especially in the subset of patients with a shockable rhythm. However survival rapidly decreased with increasing adrenaline administration time intervals (AATI).

Outbreak of cryptosporidiosis associated with a man-made chlorinated lake--Tarrant County, Texas, 2008.

In July 2008, clusters of laboratory-confirmed cryptosporidiosis cases and reports of gastrointestinal illness in persons who visited a lake were reported to Tarrant County Public Health. In response, epidemiologic, laboratory, and environmental health investigations were initiated. A matched case-control study determined that swallowing the lake water was associated with illness (adjusted odds ratio = 16.3; 95% confidence interval: 2.5-infinity). The environmental health investigation narrowed down the potential sources of contamination. Laboratory testing detected Cryptosporidium hominis in case-patient stool specimens and Cryptosporidium species in lake water. It was only through the joint effort that epidemiologic, laboratory, and environmental health investigators could determine that >1 human diarrheal fecal incidents in the lake likely led to contamination of the water. This same collaborative effort will be needed to develop and maintain an effective national Model Aquatic Health Code.

Gasping in response to basic resuscitation efforts: observation in a Swine model of cardiac arrest.

Objective. To analyze the effect of basic resuscitation efforts on gasping and of gasping on survival. Methods. This is secondary analysis of a previously reported study comparing continuous chest compressions (CCC CPR) versus chest compressions plus ventilation (30:2 CPR) on survival. 64 swine were randomized to 1 of these 2 basic CPR approaches after either short (3 or 4 minutes) or long (5 or 6 minutes) durations of untreated VF. At 12 minutes of VF, all received the same Guidelines 2005 Advanced Cardiac Life Support. Neurologically status was evaluated at 24 hours. A score of 1 is normal, 2 is abnormal, such as not eating or drinking normally, unsteady gait, or slight resistance to restraint, 3 severely abnormal, where the animal is recumbent and unable to stand, 4 is comatose, and 5 is dead. For this analysis a neurological outcome score of 1 or 2 was classified as "good", and a score of 3, 4, or 5 was classified as "poor." Results. Gasping was more likely to continue or if absent, to resume in the animals with short durations of untreated VF before basic resuscitation efforts. With long durations of untreated VF, the frequency of gasping and survival was better in swine receiving CCC CPR. In the absence of frequent gasping, intact survival was rare in the long duration of untreated VF group. Conclusions. Gasping is an important phenomenon during basic resuscitation efforts for VF arrest and in this model was more frequent with CCC-CPR.

Part 8: adult advanced cardiovascular life support: 2010 American Heart Association Guidelines for Cardiopulmonary Resuscitation and Emergency Cardiovascular Care.

The goal of therapy for bradycardia or tachycardia is to rapidly identify and treat patients who are hemodynamically unstable or symptomatic due to the arrhythmia. Drugs or, when appropriate, pacing may be used to control unstable or symptomatic bradycardia. Cardioversion or drugs or both may be used to control unstable or symptomatic tachycardia. ACLS providers should closely monitor stable patients pending expert consultation and should be prepared to aggressively treat those with evidence of decompensation.

Part 7: CPR techniques and devices: 2010 American Heart Association Guidelines for Cardiopulmonary Resuscitation and Emergency Cardiovascular Care.

A variety of CPR techniques and devices may improve hemodynamics or short-term survival when used by well-trained providers in selected patients. All of these techniques and devices have the potential to delay chest compressions and defibrillation. In order to prevent delays and maximize efficiency, initial training, ongoing monitoring, and retraining programs should be offered to providers on a frequent and ongoing basis. To date, no adjunct has consistently been shown to be superior to standard conventional (manual) CPR for out-of-hospital basic life support, and no device other than a defibrillator has consistently improved long-term survival from out-of-hospital cardiac arrest.

Continued breathing followed by gasping or apnea in a swine model of ventricular fibrillation cardiac arrest.

BACKGROUND: Continued breathing following ventricular fibrillation has here-to-fore not been described. METHODS: We analyzed the spontaneous ventilatory activity during the first several minutes of ventricular fibrillation (VF) in our isoflurane anesthesized swine model of out-of-hospital cardiac arrest. The frequency and type of ventilatory activity was monitored by pneumotachometer and main stream infrared capnometer and analyzed in 61 swine during the first 3 to 6 minutes of untreated VF. RESULTS: During the first minute of VF, the air flow pattern in all 61 swine was similar to those recorded during regular spontaneous breathing during anesthesia and was clearly different from the patterns of gasping. The average rate of continued breathing during the first minute of untreated VF was 10 breaths per minute. During the second minute of untreated VF, spontaneous breathing activity either stopped or became typical of gasping. During minutes 2 to 5 of untreated VF, most animals exhibited very slow spontaneous ventilatory activity with a pattern typical of gasping; and the pattern of gasping was crescendo-decrescendo, as has been previously reported. In the absence of therapy, all ventilatory activity stopped 6 minutes after VF cardiac arrest. CONCLUSION: In our swine model of VF cardiac arrest, we documented that normal breathing continued for the first minute following cardiac arrest.

Ocular and respiratory symptoms among lifeguards at a hotel indoor waterpark resort.

OBJECTIVES: To determine the cause of eye and respiratory irritation symptoms among lifeguards at an indoor waterpark. METHODS: Investigators 1) performed environmental sampling for chloramine, endotoxin, and microbials; 2) administered symptom questionnaires; 3) reviewed ventilation system designs; and 4) reviewed water chemistry. RESULTS: Airborne trichloramine concentrations were found at levels reported to cause irritation symptoms in other studies. Some endotoxin concentrations were found at levels associated with cough and fever in previous studies. Exposed lifeguards were significantly more likely to report work-related irritation symptoms than unexposed individuals. The ventilation system may not have provided sufficient air movement and distribution to adequately capture and remove air contaminants at deck level. No water microbes were detected, and water chemistry met state standards. CONCLUSIONS: Indoor waterparks need to control water chemistry and ensure adequate air movement and distribution to control air contaminants and reduce health symptoms.

Continuous chest compression resuscitation in arrested swine with upper airway inspiratory obstruction.

BACKGROUND: This study was designed to compare 24-h survival rates and neurological function of swine in cardiac arrest treated with one of three forms of simulated basic life support CPR. METHODS: Thirty swine were randomized equally among three experimental groups to receive either 30:2 CPR with an unobstructed endotracheal tube (ET) or continuous chest compression (CCC) CPR with an unobstructed ET or CCC CPR with a collapsable rubber sleeve on the ET allowing air outflow but completely restricting air inflow. The swine were anesthetized but not paralyzed. Two min of untreated VF was followed by 9 min of simulated single rescuer bystander CPR. In the 30:2 CPR group, each set of 30 chest compressions was followed by a 15-s pause to simulate the realistic duration of interrupted chest compressions required for a single rescuer to deliver 2 mouth-to-mouth ventilations. The other two groups were provided continuous chest compressions (CCC) without assisted ventilations. At 11 min post-arrest a biphasic defibrillation shock (150 J) was administered followed by a period of advanced cardiac life support. RESULTS: In the 30:2 group, 8 of 10 animals had good neurological function at 24-h post-resuscitation. In the CCC open airway group, 10 of 10, and in the CCC inspiratory obstructed group, 9 of 10. The number of shocks (P<0.05) and epinephrine doses (P<0.05) required for ROSC was greater in the 30:2 CPR group than in the other two groups. CONCLUSIONS: There were no differences in 24-h survival with good neurological function among these three different CPR protocols.

Shipshape: sanitation inspections on cruise ships, 1990-2005, Vessel Sanitation Program, Centers for Disease Control and Prevention.

In the course of a successful collaboration between the Centers for Disease Control and Prevention (CDC) and the cruise ship industry on reducing common-source outbreaks, CDC's Vessel Sanitation Program (VSP) has expanded its training, education, and cruise ship inspection programs. The study reported here evaluated 15 years of ship sanitation inspection data from the National Center for Environmental Health and assessed performance in specific sanitation categories from 1996 to 2005. During the period 1990-2005, scores from cruise ship environmental sanitation inspections steadily improved. The percentage of inspections with violations decreased among five of nine categories. Those five categories were Washing Facilities, Contact Surfaces, Facility Maintenance, Food Handling, and Communicable Disease Practices. Inspection violations increased proportionally in the categories of Swimming Pools and Water System Protection/Chart Recording. Overall continued good performance in most sanitation categories is likely attributable to on-site training during inspections, improvements in ship construction, and a switch from hot-holding temperatures to time limits as a public health control for foods on display.

Improved neurological outcome with continuous chest compressions compared with 30:2 compressions-to-ventilations cardiopulmonary resuscitation in a realistic swine model of out-of-hospital cardiac arrest.

BACKGROUND: The 2005 Guidelines for Cardiopulmonary Resuscitation and Emergency Cardiovascular Care changed the previous ventilations-to-chest-compression algorithm for bystander cardiopulmonary resuscitation (CPR) from 2 ventilations before each 15 chest compressions (2:15 CPR) to 30 chest compressions before 2 ventilations (30:2 CPR). It was acknowledged in the guidelines that the change was based on a consensus rather than clear evidence. This study was designed to compare 24-hour neurologically normal survival between the initial applications of continuous chest compressions without assisted ventilations with 30:2 CPR in a swine model of witnessed out-of-hospital ventricular fibrillation cardiac arrest. METHODS AND RESULTS: Sixty-four animals underwent 12 minutes of ventricular fibrillation before defibrillation attempts. They were divided into 4 groups, each with increasing durations (3, 4, 5, and 6 minutes, respectively) of untreated ventricular fibrillation before the initiation of bystander resuscitation consisting of either continuous chest compression or 30:2 CPR. After the various untreated ventricular durations plus bystander resuscitation durations, all animals were given the first defibrillation attempt 12 minutes after the induction of ventricular fibrillation, followed by the 2005 guideline-recommended advanced cardiac life support. Neurologically normal survival at 24 hours after resuscitation was observed in 23 of 33 (70%) of the animals in the continuous chest compression groups but in only 13 of 31 (42%) of the 30:2 CPR groups (P=0.025). CONCLUSIONS: In a realistic model of out-of-hospital ventricular fibrillation cardiac arrest, initial bystander administration of continuous chest compressions without assisted ventilations resulted in significantly better 24-hour postresuscitation neurologically normal survival than did the initial bystander administration of 2005 guideline-recommended 30:2 CPR.

Outbreaks of short-incubation ocular and respiratory illness following exposure to indoor swimming pools.

OBJECTIVES: Chlorination destroys pathogens in swimming pool water, but by-products of chlorination can cause human illness. We investigated outbreaks of ocular and respiratory symptoms associated with chlorinated indoor swimming pools at two hotels. MEASUREMENTS: We interviewed registered guests and companions who stayed at hotels X and Y within 2 days of outbreak onset. We performed bivariate and stratified analyses, calculated relative risks (RR), and conducted environmental investigations of indoor pool areas. RESULTS: Of 77 guests at hotel X, 47 (61%) completed questionnaires. Among persons exposed to the indoor pool area, 22 (71%) of 31 developed ocular symptoms [RR = 24; 95% confidence interval (CI), 1.5-370], and 14 (45%) developed respiratory symptoms (RR = 6.8; 95% CI, 1.0-47) with a median duration of 10 hr (0.25-24 hr). We interviewed 30 (39%) of 77 registered persons and 59 unregistered companions at hotel Y. Among persons exposed to the indoor pool area, 41 (59%) of 69 developed ocular symptoms (RR = 24; 95% CI, 1.5-370), and 28 (41%) developed respiratory symptoms (RR = 17; 95% CI, 1.1-260) with a median duration of 2.5 hr (2 min-14 days). Four persons sought medical care. During the outbreak, the hotel X's ventilation system malfunctioned. Appropriate water and air samples were not available for laboratory analysis. CONCLUSIONS AND RELEVANCE TO PROFESSIONAL PRACTICE: Indoor pool areas were associated with illness in these outbreaks. A large proportion of bathers were affected; symptoms were consistent with chloramine exposure and were sometimes severe. Improved staff training, pool maintenance, and pool area ventilation could prevent future outbreaks.