This paper proposes a novel normalizing flow learning based method IIAG-CoFlow for low-light image enhancement (LLIE), which consists of an inter-and intra-channel attention Transformer based conditional generator (IIAG) and a complete flow (CoFlow). On the one hand, IIAG is designed as a U-shape network, whose down-sampling and up-sampling layers are constructed by IIZAT (i.e., inter-and intra-channel and zero-map attention Transformer) and IIAT (i.e., inter-and intra-channel attention Transformer) respectively. IIAT is designed to calculate inter-channel attention and intra-channel attention independently. Based on IIAT, IIZAT is designed to perform parallel fusion of zero-map attention and intra-channel attention. On the other hand, based on existing normalizing flow, we bring in unconditional affine coupling layer and design 3 invertible linear transformation layers, to develop CoFlow. The height and width axes based cross attention network (HWCAN) is proposed to learn affine/linear transformation parameters for conditional feature-driven layers of CoFlow. Experiments show that IIAG-CoFlow outperforms existing SOTA LLIE methods on several benchmark low-light datasets, and real NTM images. The source codes and pre-trained models are available athttps://github.com/NJUPT-IPR-ChenTS/IIAG-CoFlow.

As bird’s-eye-view (BEV) semantic segmentation is simple-to-visualize and easy-to-handle, it has been applied in autonomous driving to provide the surrounding information to downstream tasks. Inferring BEV semantic segmentation conditioned on multi-camera-view images is a popular scheme in the community as cheap devices and real-time processing. The recent work implemented this task by learning the content and position relationship via Vision Transformer (ViT). However, its quadratic complexity confines the relationship learning only in the latent layer, leaving the scale gap to impede the representation of fine-grained objects. In view of information absorption, when representing position-related BEV features, their weighted fusion of all view feature imposes inconducive features to disturb the fusion of conducive features. To tackle these issues, we propose a novel cross-scale hierarchical Transformer with correspondence-augmented attention for semantic segmentation inference. Specifically, we devise a hierarchical framework to refine the BEV feature representation, where the last size is only half of the final segmentation. To save the computation increase caused by this hierarchical framework, we exploit the cross-scale Transformer to learn feature relationships in a reversed-aligning way, and leverage the residual connection of BEV features to facilitate information transmission between scales. We propose correspondence-augmented attention to distinguish conducive and inconducive correspondences. It is implemented in a simple yet effective way, amplifying attention scores before the Softmax operation, so that the position-view-related and the position-view-disrelated attention scores are highlighted and suppressed. Extensive experiments demonstrate that our method has state-of-the-art performance in inferring BEV semantic segmentation conditioned on multi-camera-view images.

This paper proposes a fast HSI (hue, saturation, intensity) color space and an orthogonal triangular with column pivoting (QRCP) enhancement model to tackle the large size RGB (red, green, blue) night traffic monitoring (NTM) image. Firstly, the fast HSI (FHSI) is proposed to decompose the light and color information of the RGB image, whose hue is defined as the cosine value of the included angle, instead of the included angle in HSI. The saturation of FHSI is defined as the ratio of the projection vector length and the side length of the projection equilateral triangle, and a saturation correction model is further proposed to correct color distortion of the low-light image by adjusting the saturation of FHSI. FHSI is more concise and faster than HSI. Secondly, a novel QRCP enhancement (QRCPE) model is proposed to improve the light of the low-light image by enhancing the intensity of FHSI, which first strengthens diagonal elements of QRCP, and followed by controlling the normalization of strengthened diagonal elements of QRCP. Finally, the FHSI-QRCPE based RGB image can be obtained by transforming the processed FHSI to RGB. The experimental results on NTM, SICE, ExDark, and BDD 100K databases, indicate that the proposed FHSI-QRCPE is fast and efficient to tackle low-light image enhancement.

Virtual try-on is a promising computer vision topic with a high commercial value wherein a new garment is visually worn on a person with a photo-realistic effect. Previous studies conduct their shape and content inference at one stage, employing a single-scale warping mechanism and a relatively unsophisticated content inference mechanism. These approaches have led to suboptimal results in terms of garment warping and skin reservation under challenging try-on scenarios. To address these limitations, we propose a novel virtual try-on method via progressive inference paradigm (PGVTON) that leverages a top-down inference pipeline and a general garment try-on strategy. Specifically, we propose a robust try-on parsing inference method by disentangling semantic categories and introducing consistency. Exploiting the try-on parsing as the shape guidance, we implement the garment try-on via warping-mapping-composition. To facilitate adaptation to a wide range of try-on scenarios, we adopt a covering more and selecting one warping strategy and explicitly distinguish tasks based on alignment. Additionally, we regulate StyleGAN2 to implement re-naked skin inpainting, conditioned on the target skin shape and spatial-agnostic skin features. Experiments demonstrate that our method has state-of-the-art performance under two challenging scenarios. The code will be available athttps://github.com/NerdFNY/PGVTON.

In online clothing sales, static model images only describe specific clothing statuses towards consumers. Without increasing shooting costs, it is a subject to display clothing dynamically by synthesizing a continuous image sequence between static images. This paper proposes a novel human image sequence synthesis method by pose-shape-content inference. In the condition of two reference poses, the pose is interpolated in the pose manifold controlled by a linear parameter. The interpolated pose is transferred into the end shape by AdaIN and the attention mechanism to infer target shape. Then the content in the reference image is transferred into this target shape. In the content transfer, the visual features of the human body cluster and clothing cluster are extracted, respectively. And the Sobel gradient is adopted to extract clothing texture variation. In the feature inferring, the multiscale feature-level optical flow warps source features, and style code infusion infers new region content without source features. Extensive experiments demonstrate that our method is superior in inferring clear layouts and transferring reasonable content compared to the pose transfer baselines. Moreover, our method has been verified to apply in parsing-guided image inference and dynamic display based on the pose sequence.