1

UniHCP: A Unified Model for Human-Centric Perceptions

We introduce a Unified Model for Human-Centric Perceptions (UniHCP), which can easily handle multiple distinctly defined human-centric tasks simultaneously, be trained at scale and obtains a series of SOTA performances over a wide spectrum of human-centric benchmarks.

HumanBench: Towards General Human-centric Perception with Projector Assisted Pretraining

Human-centric perceptions include a variety of vision tasks, which have widespread industrial applications, including surveillance, autonomous driving, and metaverse. It is desirable to have a general pretrain model for versatile human-centric …

Fast-MoCo: Boost Momentum-based Contrastive Learning with Combinatorial Patches

We introduce Fast-MoCo, a simple yet effective self-supervised learning method that boosts the training speed of the momentum-based contrastive learning with combinatorial patches.

Evolving Search Space for Neural Architecture Search

We introduce a neural architecture search scheme called NSE. It can progressively accommodate new search space while maintaining the previously obtained knowledge.

User-Guided Deep Anime Line Art Colorization with Conditional Adversarial Networks

We proposed a model and a dataset for accurate anime line art colorization. This model improved the visual result over the previously proposed methods.