OmniText is a training-free generalist capable of tackling diverse text image manipulation such as text insertion, editing, rescaling, repositioning, removal, and style-controlled text insertion and editing.

PRIMEdit is a zero-shot multi-instance video editing framework that uses novel probability redistribution and sampling techniques to enable faithful instance edits while preventing unintended changes in diverse video scenarios.

We present old photo modernization using multiple references by performing stylization and enhancement in a unified manner. In order to modernize old photos, we propose a novel multi-reference-based old photo modernization (MROPM) framework consisting of a network MROPM-Net and a novel synthetic data generation scheme.

We find that GN’s inferior performance against Batch normalization (BN) is caused by: unstable training performance and sensitivity to distortion, whether it comes from external noise or perturbations introduced by the regularization. In addition, we found that GN can only help the neural network training in some specific period, unlike BN, which helps the network throughout the training. To solve these issues, we propose a new normalization layer by combining the benefit of GN and BN.

We propose to improve real image denoising performance through a better learning strategy that can enable test-time adaptation on the multi-task network using two-stage learning. The first stage pre-train the network using meta-auxiliary learning to get better meta-initialization. Meanwhile, the second stage is a meta-learning strategy to fine-tune (meta-transfer learning) the network to enable test-time adaptation on real noisy images.

CISRNet is a coarse-to-fine network for compressed image super-resolution tasks. This network consists of two main subnetworks; the coarse and refinement network, where recursive and residual learning is employed within these two networks respectively.

A faster multi-shot person re-identification system with a newly proposed key frame extraction method using face features extracted from a deep learning network.

Acknowledgment: received much help from Holy Lovenia.