State Key Laboratory of Internet of Things for Smart City, University of Macau
Abstract:The core of Multi-view Stereo(MVS) is the matching process among reference and source pixels. Cost aggregation plays a significant role in this process, while previous methods focus on handling it via CNNs. This may inherit the natural limitation of CNNs that fail to discriminate repetitive or incorrect matches due to limited local receptive fields. To handle the issue, we aim to involve Transformer into cost aggregation. However, another problem may occur due to the quadratically growing computational complexity caused by Transformer, resulting in memory overflow and inference latency. In this paper, we overcome these limits with an efficient Transformer-based cost aggregation network, namely CostFormer. The Residual Depth-Aware Cost Transformer(RDACT) is proposed to aggregate long-range features on cost volume via self-attention mechanisms along the depth and spatial dimensions. Furthermore, Residual Regression Transformer(RRT) is proposed to enhance spatial attention. The proposed method is a universal plug-in to improve learning-based MVS methods.
Abstract:Significant progress has been witnessed in learning-based Multi-view Stereo (MVS) of supervised and unsupervised settings. To combine their respective merits in accuracy and completeness, meantime reducing the demand for expensive labeled data, this paper explores a novel semi-supervised setting of learning-based MVS problem that only a tiny part of the MVS data is attached with dense depth ground truth. However, due to huge variation of scenarios and flexible setting in views, semi-supervised MVS problem (Semi-MVS) may break the basic assumption in classic semi-supervised learning, that unlabeled data and labeled data share the same label space and data distribution. To handle these issues, we propose a novel semi-supervised MVS framework, namely SE-MVS. For the simple case that the basic assumption works in MVS data, consistency regularization encourages the model predictions to be consistent between original sample and randomly augmented sample via constraints on KL divergence. For further troublesome case that the basic assumption is conflicted in MVS data, we propose a novel style consistency loss to alleviate the negative effect caused by the distribution gap. The visual style of unlabeled sample is transferred to labeled sample to shrink the gap, and the model prediction of generated sample is further supervised with the label in original labeled sample. The experimental results on DTU, BlendedMVS, GTA-SFM, and Tanks\&Temples datasets show the superior performance of the proposed method. With the same settings in backbone network, our proposed SE-MVS outperforms its fully-supervised and unsupervised baselines.
Abstract:Sequence ordering of word vector matters a lot to text reading, which has been proven in natural language processing (NLP). However, the rule of different sequence ordering in computer vision (CV) was not well explored, e.g., why the "zigzag" flattening (ZF) is commonly utilized as a default option to get the image patches ordering in vision transformers (ViTs). Notably, when decomposing multi-scale images, the ZF could not maintain the invariance of feature point positions. To this end, we investigate the Hilbert fractal flattening (HF) as another method for sequence ordering in CV and contrast it against ZF. The HF has proven to be superior to other curves in maintaining spatial locality, when performing multi-scale transformations of dimensional space. And it can be easily plugged into most deep neural networks (DNNs). Extensive experiments demonstrate that it can yield consistent and significant performance boosts for a variety of architectures. Finally, we hope that our studies spark further research about the flattening strategy of image reading.
Abstract:Self-supervised learning has not been fully explored for point cloud analysis. Current frameworks are mainly based on point cloud reconstruction. Given only 3D coordinates, such approaches tend to learn local geometric structures and contours, while failing in understanding high level semantic content. Consequently, they achieve unsatisfactory performance in downstream tasks such as classification, segmentation, etc. To fill this gap, we propose a generic Contour-Perturbed Reconstruction Network (CP-Net), which can effectively guide self-supervised reconstruction to learn semantic content in the point cloud, and thus promote discriminative power of point cloud representation. First, we introduce a concise contour-perturbed augmentation module for point cloud reconstruction. With guidance of geometry disentangling, we divide point cloud into contour and content components. Subsequently, we perturb the contour components and preserve the content components on the point cloud. As a result, self supervisor can effectively focus on semantic content, by reconstructing the original point cloud from such perturbed one. Second, we use this perturbed reconstruction as an assistant branch, to guide the learning of basic reconstruction branch via a distinct dual-branch consistency loss. In this case, our CP-Net not only captures structural contour but also learn semantic content for discriminative downstream tasks. Finally, we perform extensive experiments on a number of point cloud benchmarks. Part segmentation results demonstrate that our CP-Net (81.5% of mIoU) outperforms the previous self-supervised models, and narrows the gap with the fully-supervised methods. For classification, we get a competitive result with the fully-supervised methods on ModelNet40 (92.5% accuracy) and ScanObjectNN (87.9% accuracy). The codes and models will be released afterwards.
Abstract:The stochastic differential equation (SDE)-based random process models of volatile renewable energy sources (RESs) jointly capture the evolving probability distribution and temporal correlation in continuous time. It has enabled recent studies to remarkably improve the performance of power system dynamic uncertainty quantification and optimization. However, considering the non-homogeneous random process nature of PV, there still remains a challenging question: how can a realistic and accurate SDE model for PV power be obtained that reflects its weather-dependent uncertainty in online operation, especially when high-resolution numerical weather prediction (NWP) is unavailable for many distributed plants? To fill this gap, this article finds that an accurate SDE model for PV power can be constructed by only using the cheap data from low-resolution public weather reports. Specifically, an hourly parameterized Jacobi diffusion process is constructed to recreate the temporal patterns of PV volatility during a day. Its parameters are mapped from the public weather report using an ensemble of extreme learning machines (ELMs) to reflect the varying weather conditions. The SDE model jointly captures intraday and intrahour volatility. Statistical examination based on real-world data collected in Macau shows the proposed approach outperforms a selection of state-of-the-art deep learning-based time-series forecast methods.
Abstract:Graph Convolution Network (GCN) has been successfully used for 3D human pose estimation in videos. However, it is often built on the fixed human-joint affinity, according to human skeleton. This may reduce adaptation capacity of GCN to tackle complex spatio-temporal pose variations in videos. To alleviate this problem, we propose a novel Dynamical Graph Network (DG-Net), which can dynamically identify human-joint affinity, and estimate 3D pose by adaptively learning spatial/temporal joint relations from videos. Different from traditional graph convolution, we introduce Dynamical Spatial/Temporal Graph convolution (DSG/DTG) to discover spatial/temporal human-joint affinity for each video exemplar, depending on spatial distance/temporal movement similarity between human joints in this video. Hence, they can effectively understand which joints are spatially closer and/or have consistent motion, for reducing depth ambiguity and/or motion uncertainty when lifting 2D pose to 3D pose. We conduct extensive experiments on three popular benchmarks, e.g., Human3.6M, HumanEva-I, and MPI-INF-3DHP, where DG-Net outperforms a number of recent SOTA approaches with fewer input frames and model size.
Abstract:Vision transformers (ViTs) have been an alternative design paradigm to convolutional neural networks (CNNs). However, the training of ViTs is much harder than CNNs, as it is sensitive to the training parameters, such as learning rate, optimizer and warmup epoch. The reasons for training difficulty are empirically analysed in ~\cite{xiao2021early}, and the authors conjecture that the issue lies with the \textit{patchify-stem} of ViT models and propose that early convolutions help transformers see better. In this paper, we further investigate this problem and extend the above conclusion: only early convolutions do not help for stable training, but the scaled ReLU operation in the \textit{convolutional stem} (\textit{conv-stem}) matters. We verify, both theoretically and empirically, that scaled ReLU in \textit{conv-stem} not only improves training stabilization, but also increases the diversity of patch tokens, thus boosting peak performance with a large margin via adding few parameters and flops. In addition, extensive experiments are conducted to demonstrate that previous ViTs are far from being well trained, further showing that ViTs have great potential to be a better substitute of CNNs.
Abstract:Self-supervised Multi-view stereo (MVS) with a pretext task of image reconstruction has achieved significant progress recently. However, previous methods are built upon intuitions, lacking comprehensive explanations about the effectiveness of the pretext task in self-supervised MVS. To this end, we propose to estimate epistemic uncertainty in self-supervised MVS, accounting for what the model ignores. Specially, the limitations can be categorized into two types: ambiguious supervision in foreground and invalid supervision in background. To address these issues, we propose a novel Uncertainty reduction Multi-view Stereo (UMVS) framework for self-supervised learning. To alleviate ambiguous supervision in foreground, we involve extra correspondence prior with a flow-depth consistency loss. The dense 2D correspondence of optical flows is used to regularize the 3D stereo correspondence in MVS. To handle the invalid supervision in background, we use Monte-Carlo Dropout to acquire the uncertainty map and further filter the unreliable supervision signals on invalid regions. Extensive experiments on DTU and Tank&Temples benchmark show that our U-MVS framework achieves the best performance among unsupervised MVS methods, with competitive performance with its supervised opponents.
Abstract:Face recognition has achieved significant progress in deep-learning era due to the ultra-large-scale and well-labeled datasets. However, training on ultra-large-scale datasets is time-consuming and takes up a lot of hardware resource. Therefore, designing an efficient training approach is crucial and indispensable. The heavy computational and memory costs mainly result from the high dimensionality of the Fully-Connected (FC) layer. Specifically, the dimensionality is determined by the number of face identities, which can be million-level or even more. To this end, we propose a novel training approach for ultra-large-scale face datasets, termed Faster Face Classification (F$^2$C). In F$^2$C, we first define a Gallery Net and a Probe Net that are used to generate identities' centers and extract faces' features for face recognition, respectively. Gallery Net has the same structure as Probe Net and inherits the parameters from Probe Net with a moving average paradigm. After that, to reduce the training time and hardware costs of the FC layer, we propose a Dynamic Class Pool (DCP) that stores the features from Gallery Net and calculates the inner product (logits) with positive samples (whose identities are in the DCP) in each mini-batch. DCP can be regarded as a substitute for the FC layer but it is far smaller, thus greatly reducing the computational and memory costs. For negative samples (whose identities are not in DCP), we minimize the cosine similarities between negative samples and those in DCP. Then, to improve the update efficiency of DCP's parameters, we design a dual data-loader including identity-based and instance-based loaders to generate a certain of identities and samples in mini-batches.
Abstract:We present 3 different question-answering models trained on the SQuAD2.0 dataset -- BIDAF, DocumentQA and ALBERT Retro-Reader -- demonstrating the improvement of language models in the past three years. Through our research in fine-tuning pre-trained models for question-answering, we developed a novel approach capable of achieving a 2% point improvement in SQuAD2.0 F1 in reduced training time. Our method of re-initializing select layers of a parameter-shared language model is simple yet empirically powerful.