1 code implementation • 14 Oct 2022 • Van-Anh Nguyen, Khanh Pham Dinh, Long Tung Vuong, Thanh-Toan Do, Quan Hung Tran, Dinh Phung, Trung Le
Our approach departs from the computational process of ViTs with a focus on visualizing the local and global information in input images and the latent feature embeddings at multiple levels.
2 code implementations • 19 Sep 2022 • Chuanxia Zheng, Long Tung Vuong, Jianfei Cai, Dinh Phung
Although two-stage Vector Quantized (VQ) generative models allow for synthesizing high-fidelity and high-resolution images, their quantization operator encodes similar patches within an image into the same index, resulting in a repeated artifact for similar adjacent regions using existing decoder architectures.
no code implementations • 29 Sep 2021 • Long Tung Vuong, Trung Quoc Phung, Toan Tran, Anh Tuan Tran, Dinh Phung, Trung Le
To achieve a satisfactory generalization performance on prediction tasks in an unseen domain, existing domain generalization (DG) approaches often rely on the strict assumption of fixed domain-invariant features and common hypotheses learned from a set of training domains.