Sheng Liu

Xiao Lu, Yihong Cao, Sheng Liu et al.

It is challenging to annotate large-scale datasets for supervised video shadow detection methods. Using a model trained on labeled images to the video frames directly may lead to high generalization error and temporal inconsistent results. In this paper, we address these challenges by proposing a Spatio-Temporal Interpolation Consistency Training (STICT) framework to rationally feed the unlabeled video frames together with the labeled images into an image shadow detection network training. Specifically, we propose the Spatial and Temporal ICT, in which we define two new interpolation schemes, \textit{i.e.}, the spatial interpolation and the temporal interpolation. We then derive the spatial and temporal interpolation consistency constraints accordingly for enhancing generalization in the pixel-wise classification task and for encouraging temporal consistent predictions, respectively. In addition, we design a Scale-Aware Network for multi-scale shadow knowledge learning in images, and propose a scale-consistency constraint to minimize the discrepancy among the predictions at different scales. Our proposed approach is extensively validated on the ViSha dataset and a self-annotated dataset. Experimental results show that, even without video labels, our approach is better than most state of the art supervised, semi-supervised or unsupervised image/video shadow detection methods and other methods in related tasks. Code and dataset are available at \url{https://github.com/yihong-97/STICT}.

Sheng Liu, Xiaohan Nie, Raffay Hamid

Existing approaches for Structure from Motion (SfM) produce impressive 3-D reconstruction results especially when using imagery captured with large parallax. However, to create engaging video-content in movies and TV shows, the amount by which a camera can be moved while filming a particular shot is often limited. The resulting small-motion parallax between video frames makes standard geometry-based SfM approaches not as effective for movies and TV shows. To address this challenge, we propose a simple yet effective approach that uses single-frame depth-prior obtained from a pretrained network to significantly improve geometry-based SfM for our small-parallax setting. To this end, we first use the depth-estimates of the detected keypoints to reconstruct the point cloud and camera-pose for initial two-view reconstruction. We then perform depth-regularized optimization to register new images and triangulate the new points during incremental reconstruction. To comprehensively evaluate our approach, we introduce a new dataset (StudioSfM) consisting of 130 shots with 21K frames from 15 studio-produced videos that are manually annotated by a professional CG studio. We demonstrate that our approach: (a) significantly improves the quality of 3-D reconstruction for our small-parallax setting, (b) does not cause any degradation for data with large-parallax, and (c) maintains the generalizability and scalability of geometry-based sparse SfM. Our dataset can be obtained at https://github.com/amazon-research/small-baseline-camera-tracking.

Huaxi Huang, Hui Kang, Sheng Liu et al.

Convolutional Neural Networks (CNNs) have demonstrated superiority in learning patterns, but are sensitive to label noises and may overfit noisy labels during training. The early stopping strategy averts updating CNNs during the early training phase and is widely employed in the presence of noisy labels. Motivated by biological findings that the amplitude spectrum (AS) and phase spectrum (PS) in the frequency domain play different roles in the animal's vision system, we observe that PS, which captures more semantic information, can increase the robustness of DNNs to label noise, more so than AS can. We thus propose early stops at different times for AS and PS by disentangling the features of some layer(s) into AS and PS using Discrete Fourier Transform (DFT) during training. Our proposed Phase-AmplituDe DisentangLed Early Stopping (PADDLES) method is shown to be effective on both synthetic and real-world label-noise datasets. PADDLES outperforms other early stopping methods and obtains state-of-the-art performance.

Ruiyuan Lin, Sheng Liu, Jun Jiang et al.

Recovering unknown, missing, damaged, distorted, or lost information in DCT coefficients is a common task in multiple applications of digital image processing, including image compression, selective image encryption, and image communication. This paper investigates the recovery of sign bits in DCT coefficients of digital images, by proposing two different approximation methods to solve a mixed integer linear programming (MILP) problem, which is NP-hard in general. One method is a relaxation of the MILP problem to a linear programming (LP) problem, and the other splits the original MILP problem into some smaller MILP problems and an LP problem. We considered how the proposed methods can be applied to JPEG-encoded images and conducted extensive experiments to validate their performances. The experimental results showed that the proposed methods outperformed other existing methods by a substantial margin, both according to objective quality metrics and our subjective evaluation.

Sheng Liu, Cong Phuoc Huynh, Cong Chen et al.

We present a simple yet effective self-supervised pre-training method for image harmonization which can leverage large-scale unannotated image datasets. To achieve this goal, we first generate pre-training data online with our Label-Efficient Masked Region Transform (LEMaRT) pipeline. Given an image, LEMaRT generates a foreground mask and then applies a set of transformations to perturb various visual attributes, e.g., defocus blur, contrast, saturation, of the region specified by the generated mask. We then pre-train image harmonization models by recovering the original image from the perturbed image. Secondly, we introduce an image harmonization model, namely SwinIH, by retrofitting the Swin Transformer [27] with a combination of local and global self-attention mechanisms. Pre-training SwinIH with LEMaRT results in a new state of the art for image harmonization, while being label-efficient, i.e., consuming less annotated data for fine-tuning than existing methods. Notably, on iHarmony4 dataset [8], SwinIH outperforms the state of the art, i.e., SCS-Co [16] by a margin of 0.4 dB when it is fine-tuned on only 50% of the training data, and by 1.0 dB when it is trained on the full training dataset.

Yuan Feng, Yaojun Hu, Pengfei Fang et al.

This work studies the joint rain and haze removal problem. In real-life scenarios, rain and haze, two often co-occurring common weather phenomena, can greatly degrade the clarity and quality of the scene images, leading to a performance drop in the visual applications, such as autonomous driving. However, jointly removing the rain and haze in scene images is ill-posed and challenging, where the existence of haze and rain and the change of atmosphere light, can both degrade the scene information. Current methods focus on the contamination removal part, thus ignoring the restoration of the scene information affected by the change of atmospheric light. We propose a novel deep neural network, named Asymmetric Dual-decoder U-Net (ADU-Net), to address the aforementioned challenge. The ADU-Net produces both the contamination residual and the scene residual to efficiently remove the rain and haze while preserving the fidelity of the scene information. Extensive experiments show our work outperforms the existing state-of-the-art methods by a considerable margin in both synthetic data and real-world data benchmarks, including RainCityscapes, BID Rain, and SPA-Data. For instance, we improve the state-of-the-art PSNR value by 2.26/4.57 on the RainCityscapes/SPA-Data, respectively. Codes will be made available freely to the research community.

Zi'an Xu, Yin Dai, Fayu Liu et al.

Background and objective: Parotid gland tumors account for approximately 2% to 10% of head and neck tumors. Preoperative tumor localization, differential diagnosis, and subsequent selection of appropriate treatment for parotid gland tumors are critical. However, the relative rarity of these tumors and the highly dispersed tissue types have left an unmet need for a subtle differential diagnosis of such neoplastic lesions based on preoperative radiomics. Recently, deep learning methods have developed rapidly, especially Transformer beats the traditional convolutional neural network in computer vision. Many new Transformer-based networks have been proposed for computer vision tasks. Methods: In this study, multicenter multimodal parotid gland MR images were collected. The Swin-Unet which was based on Transformer was used. MR images of short time inversion recovery, T1-weighted and T2-weighted modalities were combined into three-channel data to train the network. We achieved segmentation of the region of interest for parotid gland and tumor. Results: The Dice-Similarity Coefficient of the model on the test set was 88.63%, Mean Pixel Accuracy was 99.31%, Mean Intersection over Union was 83.99%, and Hausdorff Distance was 3.04. Then a series of comparison experiments were designed in this paper to further validate the segmentation performance of the algorithm. Conclusions: Experimental results showed that our method has good results for parotid gland and tumor segmentation. The Transformer-based network outperforms the traditional convolutional neural network in the field of medical images.

Sheng Liu, Kangning Liu, Weicheng Zhu et al.

Deep learning in the presence of noisy annotations has been studied extensively in classification, but much less in segmentation tasks. In this work, we study the learning dynamics of deep segmentation networks trained on inaccurately-annotated data. We discover a phenomenon that has been previously reported in the context of classification: the networks tend to first fit the clean pixel-level labels during an "early-learning" phase, before eventually memorizing the false annotations. However, in contrast to classification, memorization in segmentation does not arise simultaneously for all semantic categories. Inspired by these findings, we propose a new method for segmentation from noisy annotations with two key elements. First, we detect the beginning of the memorization phase separately for each category during training. This allows us to adaptively correct the noisy annotations in order to exploit early learning. Second, we incorporate a regularization term that enforces consistency across scales to boost robustness against annotation noise. Our method outperforms standard approaches on a medical-imaging segmentation task where noises are synthesized to mimic human annotation errors. It also provides robustness to realistic noisy annotations present in weakly-supervised semantic segmentation, achieving state-of-the-art results on PASCAL VOC 2012. Code is available at https://github.com/Kangningthu/ADELE

Xiaozhen Xie, Sheng Liu

Low-rankness is important in the hyperspectral image (HSI) denoising tasks. The tensor nuclear norm (TNN), defined based on the tensor singular value decomposition, is a state-of-the-art method to describe the low-rankness of HSI. However, TNN ignores some physical meanings of HSI in tackling denoising tasks, leading to suboptimal denoising performance. In this paper, we propose the multi-modal and frequency-weighted tensor nuclear norm (MFWTNN) and the non-convex MFWTNN for HSI denoising tasks. Firstly, we investigate the physical meaning of frequency slices and reconsider their weights to improve the low-rank representation ability of TNN. Secondly, we consider the correlation among two spatial dimensions and the spectral dimension of HSI and combine the above improvements to TNN to propose MFWTNN. Thirdly, we use non-convex functions to approximate the rank function of the frequency tensor and propose the NonMFWTNN to relax the MFWTNN better. Besides, we adaptively choose bigger weights for slices mainly containing noise information and smaller weights for slices containing profile information. Finally, we develop the efficient alternating direction method of multiplier (ADMM) based algorithm to solve the proposed models, and the effectiveness of our models are substantiated in simulated and real HSI datasets.

Sheng Liu, Xiao Li, Yuexiang Zhai et al.

Normalization techniques have become a basic component in modern convolutional neural networks (ConvNets). In particular, many recent works demonstrate that promoting the orthogonality of the weights helps train deep models and improve robustness. For ConvNets, most existing methods are based on penalizing or normalizing weight matrices derived from concatenating or flattening the convolutional kernels. These methods often destroy or ignore the benign convolutional structure of the kernels; therefore, they are often expensive or impractical for deep ConvNets. In contrast, we introduce a simple and efficient "Convolutional Normalization" (ConvNorm) method that can fully exploit the convolutional structure in the Fourier domain and serve as a simple plug-and-play module to be conveniently incorporated into any ConvNets. Our method is inspired by recent work on preconditioning methods for convolutional sparse coding and can effectively promote each layer's channel-wise isometry. Furthermore, we show that our ConvNorm can reduce the layerwise spectral norm of the weight matrices and hence improve the Lipschitzness of the network, leading to easier training and improved robustness for deep ConvNets. Applied to classification under noise corruptions and generative adversarial network (GAN), we show that the ConvNorm improves the robustness of common ConvNets such as ResNet and the performance of GAN. We verify our findings via numerical experiments on CIFAR and ImageNet.

Sheng Liu, Xiaozhen Xie, Wenfeng Kong

Hyperspectral images (HSIs) usually suffer from different types of pollution. This severely reduces the quality of HSIs and limits the accuracy of subsequent processing tasks. HSI denoising can be modeled as a low-rank tensor denoising problem. Tensor nuclear norm (TNN) induced by tensor singular value decomposition plays an important role in this problem. In this letter, we first reconsider three inconspicuous but crucial phenomenons in TNN. In the Fourier transform domain of HSIs, different frequency slices (FS) contain different information; different singular values (SVs) of each FS also represent different information. The two physical phenomenons lie not only in the spectral mode but also in the spatial modes. Then based on them, we propose a multi-modal and double-weighted TNN. It can adaptively shrink the FS and SVs according to their physical meanings in all modes of HSIs. In the framework of the alternating direction method of multipliers, we design an effective alternating iterative strategy to optimize our proposed model. Denoised experiments on both synthetic and real HSI datasets demonstrate their superiority against related methods.

Hao Yang, ZhongLiang Yang, YongJian Bao et al.

Extracting correlation features between codes-words with high computational efficiency is crucial to steganalysis of Voice over IP (VoIP) streams. In this paper, we utilized attention mechanisms, which have recently attracted enormous interests due to their highly parallelizable computation and flexibility in modeling correlation in sequence, to tackle steganalysis problem of Quantization Index Modulation (QIM) based steganography in compressed VoIP stream. We design a light-weight neural network named Fast Correlation Extract Model (FCEM) only based on a variant of attention called multi-head attention to extract correlation features from VoIP frames. Despite its simple form, FCEM outperforms complicated Recurrent Neural Networks (RNNs) and Convolutional Neural Networks (CNNs) models on both prediction accuracy and time efficiency. It significantly improves the best result in detecting both low embedded rates and short samples recently. Besides, the proposed model accelerates the detection speed as twice as before when the sample length is as short as 0.1s, making it a excellent method for online services.