Xinyu Zhang

Zhiyin Shao, Xinyu Zhang, Meng Fang et al.

Text-to-image person re-identification (ReID) aims to search for pedestrian images of an interested identity via textual descriptions. It is challenging due to both rich intra-modal variations and significant inter-modal gaps. Existing works usually ignore the difference in feature granularity between the two modalities, i.e., the visual features are usually fine-grained while textual features are coarse, which is mainly responsible for the large inter-modal gaps. In this paper, we propose an end-to-end framework based on transformers to learn granularity-unified representations for both modalities, denoted as LGUR. LGUR framework contains two modules: a Dictionary-based Granularity Alignment (DGA) module and a Prototype-based Granularity Unification (PGU) module. In DGA, in order to align the granularities of two modalities, we introduce a Multi-modality Shared Dictionary (MSD) to reconstruct both visual and textual features. Besides, DGA has two important factors, i.e., the cross-modality guidance and the foreground-centric reconstruction, to facilitate the optimization of MSD. In PGU, we adopt a set of shared and learnable prototypes as the queries to extract diverse and semantically aligned features for both modalities in the granularity-unified feature space, which further promotes the ReID performance. Comprehensive experiments show that our LGUR consistently outperforms state-of-the-arts by large margins on both CUHK-PEDES and ICFG-PEDES datasets. Code will be released at https://github.com/ZhiyinShao-H/LGUR.

Yan Gong, Xinyu Zhang, Hao Liu et al.

Multi-modal fusion is increasingly being used for autonomous driving tasks, as different modalities provide unique information for feature extraction. However, the existing two-stream networks are only fused at a specific network layer, which requires a lot of manual attempts to set up. As the CNN goes deeper, the two modal features become more and more advanced and abstract, and the fusion occurs at the feature level with a large gap, which can easily hurt the performance. To reduce the loss of height and depth information during the process of projecting point clouds into 2D space, we utilize calibration parameters to project the point cloud into Altitude Difference Images (ADIs), which exhibit more distinct road features. In this study, we propose a novel fusion architecture called Skip-cross Networks (SkipcrossNets), which combine adaptively ADIs and camera images without being bound to a certain fusion epoch. Specifically, skip-cross fusion strategy connects each layer to each layer in a feed-forward manner, and for each layer, the feature maps of all previous layers are used as input and its own feature maps are used as input to all subsequent layers for the other modality, enhancing feature propagation and multi-modal features fusion. This strategy facilitates selection of the most similar feature layers from two modalities, enhancing feature reuse and providing complementary effects for sparse point cloud features. The advantages of skip-cross fusion strategy is demonstrated through application to the KITTI and A2D2 datasets, achieving a MaxF score of 96.85% on KITTI and an F1 score of 84.84% on A2D2. The model parameters require only 2.33 MB of memory at a speed of 68.24 FPS, which can be viable for mobile terminals and embedded devices.

Liang Zhao, Xinyuan Zhao, Hailong Ma et al.

Most existing image inpainting algorithms are based on a single view, struggling with large holes or the holes containing complicated scenes. Some reference-guided algorithms fill the hole by referring to another viewpoint image and use 2D image alignment. Due to the camera imaging process, simple 2D transformation is difficult to achieve a satisfactory result. In this paper, we propose 3DFill, a simple and efficient method for reference-guided image inpainting. Given a target image with arbitrary hole regions and a reference image from another viewpoint, the 3DFill first aligns the two images by a two-stage method: 3D projection + 2D transformation, which has better results than 2D image alignment. The 3D projection is an overall alignment between images and the 2D transformation is a local alignment focused on the hole region. The entire process of image alignment is self-supervised. We then fill the hole in the target image with the contents of the aligned image. Finally, we use a conditional generation network to refine the filled image to obtain the inpainting result. 3DFill achieves state-of-the-art performance on image inpainting across a variety of wide view shifts and has a faster inference speed than other inpainting models.

Zhiyin Shao, Xinyu Zhang, Changxing Ding et al.

The pre-training task is indispensable for the text-to-image person re-identification (T2I-ReID) task. However, there are two underlying inconsistencies between these two tasks that may impact the performance; i) Data inconsistency. A large domain gap exists between the generic images/texts used in public pre-trained models and the specific person data in the T2I-ReID task. This gap is especially severe for texts, as general textual data are usually unable to describe specific people in fine-grained detail. ii) Training inconsistency. The processes of pre-training of images and texts are independent, despite cross-modality learning being critical to T2I-ReID. To address the above issues, we present a new unified pre-training pipeline (UniPT) designed specifically for the T2I-ReID task. We first build a large-scale text-labeled person dataset "LUPerson-T", in which pseudo-textual descriptions of images are automatically generated by the CLIP paradigm using a divide-conquer-combine strategy. Benefiting from this dataset, we then utilize a simple vision-and-language pre-training framework to explicitly align the feature space of the image and text modalities during pre-training. In this way, the pre-training task and the T2I-ReID task are made consistent with each other on both data and training levels. Without the need for any bells and whistles, our UniPT achieves competitive Rank-1 accuracy of, ie, 68.50%, 60.09%, and 51.85% on CUHK-PEDES, ICFG-PEDES and RSTPReid, respectively. Both the LUPerson-T dataset and code are available at https;//github.com/ZhiyinShao-H/UniPT.

Heyi Li, Jinlong Liu, Xinyu Zhang et al.

Despite recent advances in semantic manipulation using StyleGAN, semantic editing of real faces remains challenging. The gap between the $W$ space and the $W$+ space demands an undesirable trade-off between reconstruction quality and editing quality. To solve this problem, we propose to expand the latent space by replacing fully-connected layers in the StyleGAN's mapping network with attention-based transformers. This simple and effective technique integrates the aforementioned two spaces and transforms them into one new latent space called $W$++. Our modified StyleGAN maintains the state-of-the-art generation quality of the original StyleGAN with moderately better diversity. But more importantly, the proposed $W$++ space achieves superior performance in both reconstruction quality and editing quality. Despite these significant advantages, our $W$++ space supports existing inversion algorithms and editing methods with only negligible modifications thanks to its structural similarity with the $W/W$+ space. Extensive experiments on the FFHQ dataset prove that our proposed $W$++ space is evidently more preferable than the previous $W/W$+ space for real face editing. The code is publicly available for research purposes at https://github.com/AnonSubm2021/TransStyleGAN.

Xinyu Zhang, Jiewei Cao, Chunhua Shen et al.

Person re-identification (Re-ID) has achieved great improvement with deep learning and a large amount of labelled training data. However, it remains a challenging task for adapting a model trained in a source domain of labelled data to a target domain of only unlabelled data available. In this work, we develop a self-training method with progressive augmentation framework (PAST) to promote the model performance progressively on the target dataset. Specially, our PAST framework consists of two stages, namely, conservative stage and promoting stage. The conservative stage captures the local structure of target-domain data points with triplet-based loss functions, leading to improved feature representations. The promoting stage continuously optimizes the network by appending a changeable classification layer to the last layer of the model, enabling the use of global information about the data distribution. Importantly, we propose a new self-training strategy that progressively augments the model capability by adopting conservative and promoting stages alternately. Furthermore, to improve the reliability of selected triplet samples, we introduce a ranking-based triplet loss in the conservative stage, which is a label-free objective function basing on the similarities between data pairs. Experiments demonstrate that the proposed method achieves state-of-the-art person Re-ID performance under the unsupervised cross-domain setting. Code is available at: https://tinyurl.com/PASTReID

Junhui Yin, Xinyu Zhang, Lin Wu et al.

Current pre-trained vision-language models, such as CLIP, have demonstrated remarkable zero-shot generalization capabilities across various downstream tasks. However, their performance significantly degrades when test inputs exhibit different distributions. In this paper, we explore the concept of test-time prompt tuning (TTPT), which facilitates the adaptation of the CLIP model to novel downstream tasks through a one-step unsupervised optimization that involves only test samples. Inspired by in-context learning in natural language processing (NLP), we propose In-Context Prompt Learning (InCPL) for test-time visual recognition tasks, which empowers a pre-trained vision-language model with labeled examples as context information on downstream task. Specifically, InCPL associates a new test sample with very few labeled examples (sometimes just one) as context information, enabling reliable label estimation for the test sample and facilitating model adaptation. To achieve this, InCPL employs an efficient language-to-vision translator to explore the textual prior information for visual prompt learning. Further, we introduce a context-aware unsupervised loss to optimize visual prompts tailored to test samples. Finally, we design a cyclic learning strategy for visual and textual prompts to ensure mutual synergy across different modalities. This enables a pre-trained, frozen CLIP model to adapt to any task using its learned adaptive prompt. Our method demonstrates superior performance and achieves state-of-the-art results across various downstream datasets.

Jintao Rong, Xin Xie, Xinyi Yu et al.

Distilled video generation models offer fast and efficient synthesis but struggle with motion customization when guided by reference videos, especially under training-free settings. Existing training-free methods, originally designed for standard diffusion models, fail to generalize due to the accelerated generative process and large denoising steps in distilled models. To address this, we propose MotionEcho, a novel training-free test-time distillation framework that enables motion customization by leveraging diffusion teacher forcing. Our approach uses high-quality, slow teacher models to guide the inference of fast student models through endpoint prediction and interpolation. To maintain efficiency, we dynamically allocate computation across timesteps according to guidance needs. Extensive experiments across various distilled video generation models and benchmark datasets demonstrate that our method significantly improves motion fidelity and generation quality while preserving high efficiency. Project page: https://euminds.github.io/motionecho/

Xinyu Zhang, Yang Zhao, Hao Zhang

A wealth of angle problems occur when facial recognition is performed: At present, the feature extraction network presents eigenvectors with large differences between the frontal face and profile face recognition of the same person in many cases. For this reason, the state-of-the-art facial recognition network will use multiple samples for the same target to ensure that eigenvector differences caused by angles are ignored during training. However, there is another solution available, which is to generate frontal face images with profile face images before recognition. In this paper, we proposed a method of generating frontal faces with image-to-image profile faces based on Generative Adversarial Network (GAN).

Xinyu Zhang, Qiang Wang, Jian Zhang et al.

Data augmentation (DA) has been widely utilized to improve generalization in training deep neural networks. Recently, human-designed data augmentation has been gradually replaced by automatically learned augmentation policy. Through finding the best policy in well-designed search space of data augmentation, AutoAugment can significantly improve validation accuracy on image classification tasks. However, this approach is not computationally practical for large-scale problems. In this paper, we develop an adversarial method to arrive at a computationally-affordable solution called Adversarial AutoAugment, which can simultaneously optimize target related object and augmentation policy search loss. The augmentation policy network attempts to increase the training loss of a target network through generating adversarial augmentation policies, while the target network can learn more robust features from harder examples to improve the generalization. In contrast to prior work, we reuse the computation in target network training for policy evaluation, and dispense with the retraining of the target network. Compared to AutoAugment, this leads to about 12x reduction in computing cost and 11x shortening in time overhead on ImageNet. We show experimental results of our approach on CIFAR-10/CIFAR-100, ImageNet, and demonstrate significant performance improvements over state-of-the-art. On CIFAR-10, we achieve a top-1 test error of 1.36%, which is the currently best performing single model. On ImageNet, we achieve a leading performance of top-1 accuracy 79.40% on ResNet-50 and 80.00% on ResNet-50-D without extra data.