Test-Time Adaptation for Optical Flow Estimation Using Motion Vectors.

Due to the prohibitive cost as well as technical challenges in annotating ground-truth optical flow for large-scale realistic video datasets, the existing deep learning models for optical flow estimation mostly rely on synthetic data for training, which in turn may lead to significant performance degradation under test-data distribution shift in real-world environments. In this work, we propose the methodology to tackle this important problem. We design a self-supervised learning task for adjusting the optical flow estimation model at test time. We exploit the fact that most videos are stored in compressed formats, from which compact information on motion, in the form of motion vectors and residuals, can be made readily available. We formulate the self-supervised task as motion vector prediction, and link this task to optical flow estimation. To the best of our knowledge, our Test-Time Adaption guided with Motion Vectors (TTA-MV), is the first work to perform such adaptation for optical flow. The experimental results demonstrate that TTA-MV can improve the generalization capability of various well-known deep learning methods for optical flow estimation, such as FlowNet, PWCNet, and RAFT.

Prediction of COVID-19 Patients' Survival by Deep Learning Approaches.

Background: Despite many studies done to predict severe coronavirus 2019 (COVID-19) patients, there is no applicable clinical prediction model to predict and distinguish severe patients early. Based on laboratory and demographic data, we have developed and validated a deep learning model to predict survival and assist in the triage of COVID-19 patients in the early stages. Methods: This retrospective study developed a survival prediction model based on the deep learning method using demographic and laboratory data. The database consisted of data from 487 patients with COVID-19 diagnosed by the reverse transcription-polymerase chain reaction test and admitted to Imam Khomeini hospital affiliated to Tehran University of Medical Sciences from February 21, 2020, to June 24, 2020. Results: The developed model achieved an area under the curve (AUC) of 0.96 for survival prediction. The results demonstrated the developed model provided high precision (0.95, 0.93), recall (0.90,0.97), and F1-score (0.93,0.95) for low- and high-risk groups. Conclusion: The developed model is a deep learning-based, data-driven prediction tool that can predict the survival of COVID-19 patients with an AUC of 0.96. This model helps classify admitted patients into low-risk and high-risk groups and helps triage patients in the early stages.

High Frequency Detail Accentuation in CNN Image Restoration.

Given its nature of statistical inference, machine learning methods incline to downplay relatively rare events. But in many applications statistical outliers carry disproportional significance; they can, if being left without special treatment as of now, cause CNNs to perform unsatisfactorily on instances of interests. This is the reason why existing CNN image restoration methods all suffer from the problem of blurred details. To overcome this weakness, we advocate a new training methodology to sensitize the CNNs to desired events even they are atypical. Specifically for image restoration, we propose a so-called high frequency feature accentuation space that promotes image sharpness and clarity by maximally discriminating the ground truth image and the CNN-restored image in atypical but semantically important features. Then we force the restored image to agree with the ground truth image in the feature accentuation space by including an auxiliary loss term in the training process. This aims at a high degree of agreement of the two images on high frequency constructs such as sharp edges and fine textures, i.e., penalizes image blurs. The new CNN design method is implemented and tested for tasks of image super-resolution and denoising. Experimental results demonstrate the achievement of our design objective.

Predicting COVID-19 Incidence Through Analysis of Google Trends Data in Iran: Data Mining and Deep Learning Pilot Study.

BACKGROUND: The recent global outbreak of coronavirus disease (COVID-19) is affecting many countries worldwide. Iran is one of the top 10 most affected countries. Search engines provide useful data from populations, and these data might be useful to analyze epidemics. Utilizing data mining methods on electronic resources' data might provide a better insight into the COVID-19 outbreak to manage the health crisis in each country and worldwide. OBJECTIVE: This study aimed to predict the incidence of COVID-19 in Iran. METHODS: Data were obtained from the Google Trends website. Linear regression and long short-term memory (LSTM) models were used to estimate the number of positive COVID-19 cases. All models were evaluated using 10-fold cross-validation, and root mean square error (RMSE) was used as the performance metric. RESULTS: The linear regression model predicted the incidence with an RMSE of 7.562 (SD 6.492). The most effective factors besides previous day incidence included the search frequency of handwashing, hand sanitizer, and antiseptic topics. The RMSE of the LSTM model was 27.187 (SD 20.705). CONCLUSIONS: Data mining algorithms can be employed to predict trends of outbreaks. This prediction might support policymakers and health care managers to plan and allocate health care resources accordingly.