Head pose estimation from RGB images without depth information is a challenging task due to the loss of spatial information as well as large head pose variations in the wild. The performance of existing landmark-free methods remains unsatisfactory as the quality of estimated pose is inferior. In this paper, we propose a novel three-branch network architecture, termed as Feature Decoupling Network (FDN), a more powerful architecture for landmark-free head pose estimation from a single RGB image. In FDN, we first propose a feature decoupling (FD) module to explicitly learn the discriminative features for each pose angle by adaptively recalibrating its channel-wise responses. Besides, we introduce a cross-category center (CCC) loss to constrain the distribution of the latent variable subspaces and thus we can obtain more compact and distinct subspaces. Extensive experiments on both in-the-wild and controlled environment datasets demonstrate that the proposed method outperforms other state-of-the-art methods based on a single RGB image and behaves on par with approaches based on multimodal input resources.

The base-free aerobic oxidation of 5-hydroxymethylfurfural (HMF) in water to 2,5-furandicarboxylic acid (FDCA) is a sustainable upgrading process for cellulosic carbohydrates. A mesoporous NiO-supported Pt nanoparticle (ca. 3 nm) catalyst was reported, which can achieve 100% selectivity to FDCA at a full conversion of HMF without the assistance of any base under mild conditions, that is, 100 °C, 10 bar O2, and 12 h. The catalyst efficiency in terms of productivity reached 22.2 molFDCA molPt –1 h–1, which is the highest value among all the supported Pt catalysts to date in literature. Moreover, the Pt/NiO catalyst showed a remarkably stable and reusable performance during five consecutive cycling. X-ray diffraction, N2 physisorption, transmission electron microscopy, X-ray photoelectron spectroscopy, CO-adsorbed diffuse reflectance Fourier transform infrared spectroscopy, temperature-programed reduction of hydrogen, temperature-programed desorption of oxygen, and electron paramagnetic resonance techniques were used to comprehensively analyze the catalysts. It was disclosed that tailoring the reactive oxygen species in NiO is an effective way to control the initial reaction rate of HMF as well as the derived intermediates (i.e., a reflection of product distribution). This can be manipulated by varying the aging temperature during the preparation of NiO; thus, the role of different oxygen species (i.e., Oads. and Olatt.) in NiO was clarified. The interaction between the metallic Pt active sites and the mobile oxygen species was found to be critical to the excellent catalytic performance.

No abstract