State-of-the-art methods in image-to-image translation are capable of learning a mapping between source and target image. However, they are still produce artifacts and are not able to learn the high-level semantics of the original image, or the most discriminative parts between the source and the target image. To improve on this, the paper introduces an Attention Guided GAN, which can identify the discriminative parts of the image and leave the background unchanged.

We conducted a comparative study of forecasting performances for power consumption in Tetuan City using Sequential Networks LSTMs and a basic Transformer, comparing them against the traditional ARIMA model. We explored the effectiveness of Sequential Networks, particularly the basic Transformer model, in forecasting power consumption time series data for Tetuan City. We aimed to assess how these modern deep learning approaches outperform the classical ARIMA model, especially in terms of Mean Squared Error (MSE) comparison.

Facial Emotion detector coded from scratch using PyTorch and OpenCV. Emotion Detector: Data: Facial expression dataset from Kaggle. Classifiers: Custom ResNet-18 implementation Transfer Learning: Pre-trained ResNet-18 on subset of ImageNet Dataset Realtime Rendering (from WebCam): Preprocessing: OpenCV Frame Facial Detector: Haar Cascade Frontal Face Detector Results on the Test Data: Sample Realtime Detection off WebCam: Link to GitHub Repository

We implement ResNet-18 architecture from scratch using PyTorch. Implementation based off the original ResNet paper named “Deep Residual Learning for Image Recognition”. We implement the ResNet-18 architecture from scratch using PyTorch and trainied it on CIFAR-10 dataset. Implementation has been inspired from the original paper named “Deep Residual Learning for Image Recognition”. Link to ResNet Paper Link to GitHub Repository

We implement 3 types of AutoEncoders on the MNIST handwritten digits dataset: Simple AutoEncoder De-Noising AutoEncoder Variational AutoEncoder Implementation of Variational AutoEncoder has been inspired from the original AutoEncoder paper named “Auto-Encoding Variational Bayes”. Here are some demonstration from the codes: Reconstruction of Simple AutoEncoder: Reconstruction of De-Noising AutoEncoder: Interpolation by Variational AutoEncoder: Link to VAE Paper