Cataract Detection Using Convolutional Neural Network with VGG-19 Model

Assessment of the Levels of Knowledge Regarding Cataract and Glaucoma in Saudi Arabia and Measurement of the Ability to Differentiate Between the Two

“…Early diagnosis of cataract may reduce visual impairment and blindness [23]. Detection of glaucoma at earlier stages is vital in preventing its progression.…”

Deepfake Detection Analyzing Hybrid Dataset Utilizing CNN and SVM

“…This accuracy score shows that the fusion of the three pre-trained models improved the results than any of the classifiers on their own. Interestingly, [24] also finds that the performance of CNN improves when testing low resolution images rather than high resolution. This model displays the capabilities of using CNN as a classifier, as well as highlighting the transferability of CNN with datasets other than the training dataset.…”

“…Researchers in [24] employ three different CNNs to classify images from the DFDC and DeepFakeTIMIT dataset. Both of these datasets are composed of videos, but the datasets are preprocessed to only include one face frame from each video.…”

“…Several image augmentations, including image cropping, random horizontal flips, gaussian blur, gaussian noise, and more were applied to the dataset in order to avoid overfitting. Three different pre-trained CNN classifiers were used in [24], with them training the last several layers of the different classifiers. Each classifier makes a prediction individually, and the average is taken from the three numbers to classify the frame as real or fake.…”

“…One limitation that plagues many models is the lack of transferability. [24] claims that models typically perform well on the dataset they're trained on, but when applied to other datasets the accuracy takes a sizable dropoff. This model aims to test this by strictly training on the DFDC dataset, but testing on the DeepFakeTIMIT, where the model achieves a 96.50% accuracy.…”