The memorization effect of deep neural networks (DNNs) plays a pivotal role in recent label noise learning methods. To exploit this effect, the model prediction-based methods have been widely adopted, which aim to exploit the outputs of DNNs in the early stage of learning to correct noisy labels. However, we observe that the model will make mistakes during label prediction, resulting in unsatisfactory performance. By contrast, the produced features in the early stage of learning show better robustness. Inspired by this observation, in this paper, we propose a novel feature embedding-based method for deep learning with label noise, termed LabEl NoiseDilution (LEND). To be specific, we first compute a similarity matrix based on current embedded features to capture the local structure of training data. Then, the noisy supervision signals carried by mislabeled data are overwhelmed by nearby correctly labeled ones (\textit{i.e.}, label noise dilution), of which the effectiveness is guaranteed by the inherent robustness of feature embedding. Finally, the training data with diluted labels are further used to train a robust classifier. Empirically, we conduct extensive experiments on both synthetic and real-world noisy datasets by comparing our LEND with several representative robust learning approaches. The results verify the effectiveness of our LEND.
The abstractive methods lack of creative ability is particularly a problem in automatic text summarization. The summaries generated by models are mostly extracted from the source articles. One of the main causes for this problem is the lack of dataset with abstractiveness, especially for Chinese. In order to solve this problem, we paraphrase the reference summaries in CLTS, the Chinese Long Text Summarization dataset, correct errors of factual inconsistencies, and propose the first Chinese Long Text Summarization dataset with a high level of abstractiveness, CLTS+, which contains more than 180K article-summary pairs and is available online. Additionally, we introduce an intrinsic metric based on co-occurrence words to evaluate the dataset we constructed. We analyze the extraction strategies used in CLTS+ summaries against other datasets to quantify the abstractiveness and difficulty of our new data and train several baselines on CLTS+ to verify the utility of it for improving the creative ability of models.
In real-world scenarios, many large-scale datasets often contain inaccurate labels, i.e., noisy labels, which may confuse model training and lead to performance degradation. To overcome this issue, Label Noise Learning (LNL) has recently attracted much attention, and various methods have been proposed to design an unbiased risk estimator to the noise-free dataset to combat such label noise. Among them, a trend of works based on Loss Decomposition and Centroid Estimation (LDCE) has shown very promising performance. However, existing LNL methods based on LDCE are only designed for binary classification, and they are not directly extendable to multi-class situations. In this paper, we propose a novel multi-class robust learning method for LDCE, which is termed "MC-LDCE". Specifically, we decompose the commonly adopted loss (e.g., mean squared loss) function into a label-dependent part and a label-independent part, in which only the former is influenced by label noise. Further, by defining a new form of data centroid, we transform the recovery problem of a label-dependent part to a centroid estimation problem. Finally, by critically examining the mathematical expectation of clean data centroid given the observed noisy set, the centroid can be estimated which helps to build an unbiased risk estimator for multi-class learning. The proposed MC-LDCE method is general and applicable to different types (i.e., linear and nonlinear) of classification models. The experimental results on five public datasets demonstrate the superiority of the proposed MC-LDCE against other representative LNL methods in tackling multi-class label noise problem.
Multi-scale representations are crucial for semantic segmentation. The community has witnessed the flourish of semantic segmentation convolutional neural networks (CNN) exploiting multi-scale contextual information. Motivated by that the vision transformer (ViT) is powerful in image classification, some semantic segmentation ViTs are recently proposed, most of them attaining impressive results but at a cost of computational economy. In this paper, we succeed in introducing multi-scale representations into semantic segmentation ViT via window attention mechanism and further improves the performance and efficiency. To this end, we introduce large window attention which allows the local window to query a larger area of context window at only a little computation overhead. By regulating the ratio of the context area to the query area, we enable the large window attention to capture the contextual information at multiple scales. Moreover, the framework of spatial pyramid pooling is adopted to collaborate with the large window attention, which presents a novel decoder named large window attention spatial pyramid pooling (LawinASPP) for semantic segmentation ViT. Our resulting ViT, Lawin Transformer, is composed of an efficient hierachical vision transformer (HVT) as encoder and a LawinASPP as decoder. The empirical results demonstrate that Lawin Transformer offers an improved efficiency compared to the existing method. Lawin Transformer further sets new state-of-the-art performance on Cityscapes (84.4\% mIoU), ADE20K (56.2\% mIoU) and COCO-Stuff datasets. The code will be released at https://github.com/yan-hao-tian/lawin.
While RGB-Infrared cross-modality person re-identification (RGB-IR ReID) has enabled great progress in 24-hour intelligent surveillance, state-of-the-arts still heavily rely on fine-tuning ImageNet pre-trained networks. Due to the single-modality nature, such large-scale pre-training may yield RGB-biased representations that hinder the performance of cross-modality image retrieval. This paper presents a self-supervised pre-training alternative, named Modality-Aware Multiple Granularity Learning (MMGL), which directly trains models from scratch on multi-modality ReID datasets, but achieving competitive results without external data and sophisticated tuning tricks. Specifically, MMGL globally maps shuffled RGB-IR images into a shared latent permutation space and further improves local discriminability by maximizing agreement between cycle-consistent RGB-IR image patches. Experiments demonstrate that MMGL learns better representations (+6.47% Rank-1) with faster training speed (converge in few hours) and solider data efficiency (<5% data size) than ImageNet pre-training. The results also suggest it generalizes well to various existing models, losses and has promising transferability across datasets. The code will be released.
Internet video delivery has undergone a tremendous explosion of growth over the past few years. However, the quality of video delivery system greatly depends on the Internet bandwidth. Deep Neural Networks (DNNs) are utilized to improve the quality of video delivery recently. These methods divide a video into chunks, and stream LR video chunks and corresponding content-aware models to the client. The client runs the inference of models to super-resolve the LR chunks. Consequently, a large number of models are streamed in order to deliver a video. In this paper, we first carefully study the relation between models of different chunks, then we tactfully design a joint training framework along with the Content-aware Feature Modulation (CaFM) layer to compress these models for neural video delivery. {\bf With our method, each video chunk only requires less than $1\% $ of original parameters to be streamed, achieving even better SR performance.} We conduct extensive experiments across various SR backbones, video time length, and scaling factors to demonstrate the advantages of our method. Besides, our method can be also viewed as a new approach of video coding. Our primary experiments achieve better video quality compared with the commercial H.264 and H.265 standard under the same storage cost, showing the great potential of the proposed method. Code is available at:\url{https://github.com/Neural-video-delivery/CaFM-Pytorch-ICCV2021}
In fringe projection profilometry, the high-order harmonics information of non-sinusoidal fringes will lead to errors in the phase estimation. In order to solve this problem, a point-wise posterior phase estimation (PWPPE) method based on deep learning technique is proposed in this paper. The complex nonlinear mapping relationship between the multiple gray values and the sine / cosine value of the phase is constructed by using the feedforward neural network model. After the model training, it can estimate the phase values of each pixel location, and the accuracy is higher than the point-wise least-square (PWLS) method. To further verify the effectiveness of this method, a face mask is measured, the traditional PWLS method and the proposed PWPPE method are employed, respectively. The comparison results show that the traditional method is with periodic phase errors, while the proposed PWPPE method can effectively eliminate such phase errors caused by non-sinusoidal fringes.
Although convolutional networks (ConvNets) have enjoyed great success in computer vision (CV), it suffers from capturing global information crucial to dense prediction tasks such as object detection and segmentation. In this work, we innovatively propose ConTNet (ConvolutionTransformer Network), combining transformer with ConvNet architectures to provide large receptive fields. Unlike the recently-proposed transformer-based models (e.g., ViT, DeiT) that are sensitive to hyper-parameters and extremely dependent on a pile of data augmentations when trained from scratch on a midsize dataset (e.g., ImageNet1k), ConTNet can be optimized like normal ConvNets (e.g., ResNet) and preserve an outstanding robustness. It is also worth pointing that, given identical strong data augmentations, the performance improvement of ConTNet is more remarkable than that of ResNet. We present its superiority and effectiveness on image classification and downstream tasks. For example, our ConTNet achieves 81.8% top-1 accuracy on ImageNet which is the same as DeiT-B with less than 40% computational complexity. ConTNet-M also outperforms ResNet50 as the backbone of both Faster-RCNN (by 2.6%) and Mask-RCNN (by 3.2%) on COCO2017 dataset. We hope that ConTNet could serve as a useful backbone for CV tasks and bring new ideas for model design