-
1.Image Quality Assessment with Gradient Siamese Network
Source: CVPR2022
Link: https://arxiv.org/abs/2208.04081
Contribution:1. Gradient Siamese Network, namely GSN.(central differential convolution (CDC), Multi-scale feature fusion);2. KL divergence loss
2. Aesthetic Attributes Assessment of Images with AMANv2 and DPC-CaptionsV2
Source:IEEE TRANSACTIONS ON MULTIMEDIA, VOL. XX, NO. X, OCTOBER 2021
Contributions: 1. Dataset DPC-CaptionsV2(four main attributes of photography: composition, lighting, color, subject) 2. propose Aesthetic Multi-Attribute Network version 2 (AMANv2)
3. Blind Image Quality Assessment Using a Deep Bilinear Convolutional Neural Network
Source: IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY, VOL. 30, NO. 1, JANUARY 2020
Contribution:1. For synthetic distortions, we construct a large-scale pre-training set based on the Waterloo Exploration Database [19] and PASCAL VOC 2012 2. For authentic distortions,we opt to use another CNN model (VGG-16 to be exact) that is pre-trained on ImageNet
4. Blind Image Quality Assessment With Active Inference
Source: IEEE TRANSACTIONS ON IMAGE PROCESSING, VOL. 30, 2021
Contribution: 1. propose a novel GAN to emulate the active inference process of IGM. proposed GAN-based active inference module can effectively predict the primary content of a distorted image for multifaceted quality analysis. 2. design a multi-stream quality evaluator network that can simultaneously measure the effects of the content-/distortion-/degradation-dependency on image quality.