How to Accelerate an Image Upscaling CNN on FPGA Using HLS

This repository contains a jupyter notebook with the implementation of a simple CNN with the objective of demonstrating how these neural networks can be used for image upscaling

Contains a module for functions that build and evaluate a U-Net Model which is a modified auto-encoder image-to-image architecture that includes skip connections from encoder to decoder layers with matching input dimensions to retain features from the encoded image. The model is designed for image segmentation and produces a mask based off x-ray data. The model is used with lung x-rays to segment out lungs, brain scans to segment out brain tumors, and pictures of a room to segment out a human figure. With the lungs dataset, it is able to manage 98% accuracy with a wide range of 3,4,5 encoding block of Conv2D and MaxPooling and BatchNormalization + 1 latent layer without pooling and the mirrored number of decoding layers that do Conv2DTranspose for deconvolution and UpScaling to reverse MaxPooling making 7,9,11 layers with doubling filters/kernels in each encoding block from 16,32,64,80 that double with each block like in a traditional CNN and then half with each decoding block.

Super-resolution using GANs. CNN, Image Classification and Image Upscaling.

Efficient sub-pixel convolutional neural networks can be used for both images and videos. ESPCN (Efficient Subpixel CNN), is a model that reconstructs a high-resolution version of an image given a low-resolution version. The ‘array’ of image upscaling filters are learned by efficient sub-pixel convolution layers.

Deep Learning in Medical Imaging (DLMI) final project. Train a CNN & GAN to upscale ultrasound images of liver cancer.

This is the computer vision project for the 5th semester

This is the computer vision project for the 5th semester

Simple CNN for image upscaling (superresolution)

Using deep cnn to upscaling satellite images

Comparing CNN models for Single-Image Super Resolution and Realtime Video Upscaling

Comparison of CNN and ViT for image upscaling and facial recognition tasks trained on racially biased datasets

General Adversarial Network and Upscaling CNN created in Pytorch to generate Anime Characters. GAN generates 64x64 pixel images which are then upscaled using ESRGAN

Повышение разрешения изображений с помощью автокодировщика и глубоких свёрточных сетей / Image upscaling using autoencoder and Convolution Neural Networks (CNN)

Designed and trained a CNN for image super-resolution for 2x-4x upscaling with a higher PSNR than interpolation, deployed as a web app: https://pixelboost.github.io

Image Super-Resolution Using CNNs, developing a deep learning model to enhance low-resolution images with improved clarity and detail. Implemented SRCNN, achieving better results than traditional upscaling methods. Built a web-based platform for real-time image enhancement, exploring AI’s potential in medical imaging, security, and digital media.

This project explores various approaches for image upscaling and classification. We trained a Super Resolution Generative Adversarial Networks (SRGAN) model and compared its performance to other algorithms such as BICUBIC, LAPSRN, and EDSR. We also evaluated the effectiveness of different CNN algorithms for image classification tasks.

An AI-powered Image Enhancement tool that improves image quality using deep learning-based super-resolution and noise reduction techniques. This project leverages CNNs and GANs to restore details, sharpen edges, and upscale low-resolution images.

This project focuses on improving the resolution of low-quality images using Convolutional Neural Networks (CNNs). The goal is to upscale images while preserving fine details, textures, and overall quality. Models were trained and tested using the

The goal is to reconstruct a high-resolution image by increasing its pixel count while maintaining or improving the image's details and sharpness. The project leverages machine learning techniques, particularly convolutional neural networks (CNNs), to train models that can upscale images with minimal distortion.

Image processing uses deep learning, like CNNs and GANs, to upscale low-resolution images by predicting missing details. Despite accuracy improvements, high-frequency details are often lacking. Our model uses a GAN framework (generator and discriminator) to produce high-quality, high-resolution images from low-quality inputs.

A U-Net-based CNN upscales images using SE blocks, pixel shuffle, and residual bottlenecks. Trained on Flickr2K with realistic degradations, MAE and perceptual-loss models improve detail recovery. Achieves strong PSNR/SSIM, producing sharper high-resolution outputs.