Search

Videos contain rich spatio-temporal information. Traditional methods for extracting motion, used in tasks such as action recognition, often rely on visual contents rather than precise motion features. This phenomenon is referred to as 'blind motion extraction' behavior, which proves inefficient in capturing motions of interest due to a lack of motion-guided cues. Recently, attention mechanisms have enhanced many computer vision tasks by effectively highlighting salient visual areas. Inspired by this, we propose using a modified Sigmoid function with learnable slope and shift parameters as an attention mechanism to activate and modulate motion signals derived from frame differencing maps. This approach generates a sequence of attention maps that enhance the processing of motion-related video content. To ensure temporally continuity and smoothness of the attention maps, we apply pair-wise temporal attention variation regularization to remove unwanted motions (e.g., noise) while preserving important ones. We then perform Hadamard product between each pair of attention maps and the original video frames to highlight the evolving motions of interest over time. These highlighted motions, termed video motion prompts, are subsequently used as inputs to the model instead of the original video frames. We formalize this process as a motion prompt layer and incorporate the regularization term into the loss function to learn better motion prompts. This layer serves as an adapter between the model and the video data, bridging the gap between traditional 'blind motion extraction' and the extraction of relevant motions of interest., Comment: Research report

Generative Adversarial Networks (GANs) significantly advanced image generation but their performance heavily depends on abundant training data. In scenarios with limited data, GANs often struggle with discriminator overfitting and unstable training. Batch Normalization (BN), despite being known for enhancing generalization and training stability, has rarely been used in the discriminator of Data-Efficient GANs. Our work addresses this gap by identifying a critical flaw in BN: the tendency for gradient explosion during the centering and scaling steps. To tackle this issue, we present CHAIN (lipsCHitz continuity constrAIned Normalization), which replaces the conventional centering step with zero-mean regularization and integrates a Lipschitz continuity constraint in the scaling step. CHAIN further enhances GAN training by adaptively interpolating the normalized and unnormalized features, effectively avoiding discriminator overfitting. Our theoretical analyses firmly establishes CHAIN's effectiveness in reducing gradients in latent features and weights, improving stability and generalization in GAN training. Empirical evidence supports our theory. CHAIN achieves state-of-the-art results in data-limited scenarios on CIFAR-10/100, ImageNet, five low-shot and seven high-resolution few-shot image datasets. Code: https://github.com/MaxwellYaoNi/CHAIN, Comment: Accepted by CVPR 2024. 26 pages. Improve Lemma 3.1 - Prop. 3.1 logic flow. Code: https://github.com/MaxwellYaoNi/CHAIN

Existing eye fixation prediction methods perform the mapping from input images to the corresponding dense fixation maps generated from raw fixation points. However, due to the stochastic nature of human fixation, the generated dense fixation maps may be a less-than-ideal representation of human fixation. To provide a robust fixation model, we introduce Gaussian Representation for eye fixation modeling. Specifically, we propose to model the eye fixation map as a mixture of probability distributions, namely a Gaussian Mixture Model. In this new representation, we use several Gaussian distribution components as an alternative to the provided fixation map, which makes the model more robust to the randomness of fixation. Meanwhile, we design our framework upon some lightweight backbones to achieve real-time fixation prediction. Experimental results on three public fixation prediction datasets (SALICON, MIT1003, TORONTO) demonstrate that our method is fast and effective., Comment: 11 pages, 7 figures

Video sequences exhibit significant nuisance variations (undesired effects) of speed of actions, temporal locations, and subjects' poses, leading to temporal-viewpoint misalignment when comparing two sets of frames or evaluating the similarity of two sequences. Thus, we propose Joint tEmporal and cAmera viewpoiNt alIgnmEnt (JEANIE) for sequence pairs. In particular, we focus on 3D skeleton sequences whose camera and subjects' poses can be easily manipulated in 3D. We evaluate JEANIE on skeletal Few-shot Action Recognition (FSAR), where matching well temporal blocks (temporal chunks that make up a sequence) of support-query sequence pairs (by factoring out nuisance variations) is essential due to limited samples of novel classes. Given a query sequence, we create its several views by simulating several camera locations. For a support sequence, we match it with view-simulated query sequences, as in the popular Dynamic Time Warping (DTW). Specifically, each support temporal block can be matched to the query temporal block with the same or adjacent (next) temporal index, and adjacent camera views to achieve joint local temporal-viewpoint warping. JEANIE selects the smallest distance among matching paths with different temporal-viewpoint warping patterns, an advantage over DTW which only performs temporal alignment. We also propose an unsupervised FSAR akin to clustering of sequences with JEANIE as a distance measure. JEANIE achieves state-of-the-art results on NTU-60, NTU-120, Kinetics-skeleton and UWA3D Multiview Activity II on supervised and unsupervised FSAR, and their meta-learning inspired fusion., Comment: Accepted by the International Journal of Computer Vision (IJCV). An extension of our ACCV'22 paper [arXiv:arXiv:2210.16820] which was distinguished by the Sang Uk Lee Best Student Paper Award

Traditional shape descriptors have been gradually replaced by convolutional neural networks due to their superior performance in feature extraction and classification. The state-of-the-art methods recognize object shapes via image reconstruction or pixel classification. However , these methods are biased toward texture information and overlook the essential shape descriptions, thus, they fail to generalize to unseen shapes. We are the first to propose a fewshot shape descriptor (FSSD) to recognize object shapes given only one or a few samples. We employ an embedding module for FSSD to extract transformation-invariant shape features. Secondly, we develop a dual attention mechanism to decompose and reconstruct the shape features via learnable shape primitives. In this way, any shape can be formed through a finite set basis, and the learned representation model is highly interpretable and extendable to unseen shapes. Thirdly, we propose a decoding module to include the supervision of shape masks and edges and align the original and reconstructed shape features, enforcing the learned features to be more shape-aware. Lastly, all the proposed modules are assembled into a few-shot shape recognition scheme. Experiments on five datasets show that our FSSD significantly improves the shape classification compared to the state-of-the-art under the few-shot setting., Comment: Accepted by WACV 2024; 8 pages for main paper

Masked Image Modeling (MIM) is a powerful self-supervised strategy for visual pre-training without the use of labels. MIM applies random crops to input images, processes them with an encoder, and then recovers the masked inputs with a decoder, which encourages the network to capture and learn structural information about objects and scenes. The intermediate feature representations obtained from MIM are suitable for fine-tuning on downstream tasks. In this paper, we propose an Image Modeling framework based on random orthogonal projection instead of binary masking as in MIM. Our proposed Random Orthogonal Projection Image Modeling (ROPIM) reduces spatially-wise token information under guaranteed bound on the noise variance and can be considered as masking entire spatial image area under locally varying masking degrees. Since ROPIM uses a random subspace for the projection that realizes the masking step, the readily available complement of the subspace can be used during unmasking to promote recovery of removed information. In this paper, we show that using random orthogonal projection leads to superior performance compared to crop-based masking. We demonstrate state-of-the-art results on several popular benchmarks., Comment: Published as a conference paper at the International Conference on Learning Representations (ICLR) 2024. 19 pages

Learning good self-supervised graph representations that are beneficial to downstream tasks is challenging. Among a variety of methods, contrastive learning enjoys competitive performance. The embeddings of contrastive learning are arranged on a hypersphere that enables the Cosine distance measurement in the Euclidean space. However, the underlying structure of many domains such as graphs exhibits highly non-Euclidean latent geometry. To this end, we propose a novel contrastive learning framework to learn high-quality graph embedding. Specifically, we design the alignment metric that effectively captures the hierarchical data-invariant information, as well as we propose a substitute of uniformity metric to prevent the so-called dimensional collapse. We show that in the hyperbolic space one has to address the leaf- and height-level uniformity which are related to properties of trees, whereas in the ambient space of the hyperbolic manifold, these notions translate into imposing an isotropic ring density towards boundaries of Poincar\'e ball. This ring density can be easily imposed by promoting the isotropic feature distribution on the tangent space of manifold. In the experiments, we demonstrate the efficacy of our proposed method across different hyperbolic graph embedding techniques in both supervised and self-supervised learning settings.

Various research studies indicate that action recognition performance highly depends on the types of motions being extracted and how accurate the human actions are represented. In this paper, we investigate different optical flow, and features extracted from these optical flow that capturing both short-term and long-term motion dynamics. We perform power normalization on the magnitude component of optical flow for flow dynamics correction to boost subtle or dampen sudden motions. We show that existing action recognition models which rely on optical flow are able to get performance boosted with our corrected optical flow. To further improve performance, we integrate our corrected flow dynamics into popular models through a simple hallucination step by selecting only the best performing optical flow features, and we show that by 'translating' the CNN feature maps into these optical flow features with different scales of motions leads to the new state-of-the-art performance on several benchmarks including HMDB-51, YUP++, fine-grained action recognition on MPII Cooking Activities, and large-scale Charades., Comment: Accepted by IEEE International Conference on Acoustics, Speech, and Signal Processing (ICASSP 2024)

In contrastive learning, two views of an original image, generated by different augmentations, are considered a positive pair, and their similarity is required to be high. Similarly, two views of distinct images form a negative pair, with encouraged low similarity. Typically, a single similarity measure, provided by a lone projection head, evaluates positive and negative sample pairs. However, due to diverse augmentation strategies and varying intra-sample similarity, views from the same image may not always be similar. Additionally, owing to inter-sample similarity, views from different images may be more akin than those from the same image. Consequently, enforcing high similarity for positive pairs and low similarity for negative pairs may be unattainable, and in some cases, such enforcement could detrimentally impact performance. To address this challenge, we propose using multiple projection heads, each producing a distinct set of features. Our pre-training loss function emerges from a solution to the maximum likelihood estimation over head-wise posterior distributions of positive samples given observations. This loss incorporates the similarity measure over positive and negative pairs, each re-weighted by an individual adaptive temperature, regulated to prevent ill solutions. Our approach, Adaptive Multi-Head Contrastive Learning (AMCL), can be applied to and experimentally enhances several popular contrastive learning methods such as SimCLR, MoCo, and Barlow Twins. The improvement remains consistent across various backbones and linear probing epochs, and becomes more significant when employing multiple augmentation methods., Comment: Accepted at the 18th European Conference on Computer Vision (ECCV 2024)

Deep learning models suffer from catastrophic forgetting when being fine-tuned with samples of new classes. This issue becomes even more pronounced when faced with the domain shift between training and testing data. In this paper, we study the critical and less explored Domain-Generalized Class-Incremental Learning (DGCIL). We design a DGCIL approach that remembers old classes, adapts to new classes, and can classify reliably objects from unseen domains. Specifically, our loss formulation maintains classification boundaries and suppresses the domain-specific information of each class. With no old exemplars stored, we use knowledge distillation and estimate old class prototype drift as incremental training advances. Our prototype representations are based on multivariate Normal distributions whose means and covariances are constantly adapted to changing model features to represent old classes well by adapting to the feature space drift. For old classes, we sample pseudo-features from the adapted Normal distributions with the help of Cholesky decomposition. In contrast to previous pseudo-feature sampling strategies that rely solely on average mean prototypes, our method excels at capturing varying semantic information. Experiments on several benchmarks validate our claims.

Traditional approaches for learning on categorical data underexploit the dependencies between columns (\aka fields) in a dataset because they rely on the embedding of data points driven alone by the classification/regression loss. In contrast, we propose a novel method for learning on categorical data with the goal of exploiting dependencies between fields. Instead of modelling statistics of features globally (i.e., by the covariance matrix of features), we learn a global field dependency matrix that captures dependencies between fields and then we refine the global field dependency matrix at the instance-wise level with different weights (so-called local dependency modelling) w.r.t. each field to improve the modelling of the field dependencies. Our algorithm exploits the meta-learning paradigm, i.e., the dependency matrices are refined in the inner loop of the meta-learning algorithm without the use of labels, whereas the outer loop intertwines the updates of the embedding matrix (the matrix performing projection) and global dependency matrix in a supervised fashion (with the use of labels). Our method is simple yet it outperforms several state-of-the-art methods on six popular dataset benchmarks. Detailed ablation studies provide additional insights into our method., Comment: IEEE Transactions on Pattern Analysis and Machine Intelligence (submitted June 2022, accepted July 2023)

We address the task of weakly-supervised few-shot image classification and segmentation, by leveraging a Vision Transformer (ViT) pretrained with self-supervision. Our proposed method takes token representations from the self-supervised ViT and leverages their correlations, via self-attention, to produce classification and segmentation predictions through separate task heads. Our model is able to effectively learn to perform classification and segmentation in the absence of pixel-level labels during training, using only image-level labels. To do this it uses attention maps, created from tokens generated by the self-supervised ViT backbone, as pixel-level pseudo-labels. We also explore a practical setup with ``mixed" supervision, where a small number of training images contains ground-truth pixel-level labels and the remaining images have only image-level labels. For this mixed setup, we propose to improve the pseudo-labels using a pseudo-label enhancer that was trained using the available ground-truth pixel-level labels. Experiments on Pascal-5i and COCO-20i demonstrate significant performance gains in a variety of supervision settings, and in particular when little-to-no pixel-level labels are available., Comment: Accepted at CVPR 2023

New roads are being constructed all the time. However, the capabilities of previous deep forecasting models to generalize to new roads not seen in the training data (unseen roads) are rarely explored. In this paper, we introduce a novel setup called a spatio-temporal (ST) split to evaluate the models' capabilities to generalize to unseen roads. In this setup, the models are trained on data from a sample of roads, but tested on roads not seen in the training data. Moreover, we also present a novel framework called Spatial Contrastive Pre-Training (SCPT) where we introduce a spatial encoder module to extract latent features from unseen roads during inference time. This spatial encoder is pre-trained using contrastive learning. During inference, the spatial encoder only requires two days of traffic data on the new roads and does not require any re-training. We also show that the output from the spatial encoder can be used effectively to infer latent node embeddings on unseen roads during inference time. The SCPT framework also incorporates a new layer, named the spatially gated addition (SGA) layer, to effectively combine the latent features from the output of the spatial encoder to existing backbones. Additionally, since there is limited data on the unseen roads, we argue that it is better to decouple traffic signals to trivial-to-capture periodic signals and difficult-to-capture Markovian signals, and for the spatial encoder to only learn the Markovian signals. Finally, we empirically evaluated SCPT using the ST split setup on four real-world datasets. The results showed that adding SCPT to a backbone consistently improves forecasting performance on unseen roads. More importantly, the improvements are greater when forecasting further into the future. The codes are available on GitHub: https://github.com/cruiseresearchgroup/forecasting-on-new-roads ., Comment: 25 pages including reference, an additional 3 pages of appendix, 8 figures. ECML PKDD 2023 Journal track special issue: Data Mining and Knowledge Discovery (DAMI)

A road network, in the context of traffic forecasting, is typically modeled as a graph where the nodes are sensors that measure traffic metrics (such as speed) at that location. Traffic forecasting is interesting because it is complex as the future speed of a road is dependent on a number of different factors. Therefore, to properly forecast traffic, we need a model that is capable of capturing all these different factors. A factor that is missing from the existing works is the node interactions factor. Existing works fail to capture the inter-node interactions because none are using the message-passing flavor of GNN, which is the one best suited to capture the node interactions This paper presents a plausible scenario in road traffic where node interactions are important and argued that the most appropriate GNN flavor to capture node interactions is message-passing. Results from real-world data show the superiority of the message-passing flavor for traffic forecasting. An additional experiment using synthetic data shows that the message-passing flavor can capture inter-node interaction better than other flavors., Comment: 18 pages, 5 figures

Few-shot learning (FSL) is popular due to its ability to adapt to novel classes. Compared with inductive few-shot learning, transductive models typically perform better as they leverage all samples of the query set. The two existing classes of methods, prototype-based and graph-based, have the disadvantages of inaccurate prototype estimation and sub-optimal graph construction with kernel functions, respectively. In this paper, we propose a novel prototype-based label propagation to solve these issues. Specifically, our graph construction is based on the relation between prototypes and samples rather than between samples. As prototypes are being updated, the graph changes. We also estimate the label of each prototype instead of considering a prototype be the class centre. On mini-ImageNet, tiered-ImageNet, CIFAR-FS and CUB datasets, we show the proposed method outperforms other state-of-the-art methods in transductive FSL and semi-supervised FSL when some unlabeled data accompanies the novel few-shot task., Comment: This paper is published at CVPR 2023

Visual representation based on covariance matrix has demonstrates its efficacy for image classification by characterising the pairwise correlation of different channels in convolutional feature maps. However, pairwise correlation will become misleading once there is another channel correlating with both channels of interest, resulting in the ``confounding'' effect. For this case, ``partial correlation'' which removes the confounding effect shall be estimated instead. Nevertheless, reliably estimating partial correlation requires to solve a symmetric positive definite matrix optimisation, known as sparse inverse covariance estimation (SICE). How to incorporate this process into CNN remains an open issue. In this work, we formulate SICE as a novel structured layer of CNN. To ensure end-to-end trainability, we develop an iterative method to solve the above matrix optimisation during forward and backward propagation steps. Our work obtains a partial correlation based deep visual representation and mitigates the small sample problem often encountered by covariance matrix estimation in CNN. Computationally, our model can be effectively trained with GPU and works well with a large number of channels of advanced CNNs. Experiments show the efficacy and superior classification performance of our deep visual representation compared to covariance matrix based counterparts., Comment: This paper is published at CVPR 2023

Unlike current deep keypoint detectors that are trained to recognize limited number of body parts, few-shot keypoint detection (FSKD) attempts to localize any keypoints, including novel or base keypoints, depending on the reference samples. FSKD requires the semantically meaningful relations for keypoint similarity learning to overcome the ubiquitous noise and ambiguous local patterns. One rescue comes with vision transformer (ViT) as it captures long-range relations well. However, ViT may model irrelevant features outside of the region of interest due to the global attention matrix, thus degrading similarity learning between support and query features. In this paper, we present a novel saliency-guided vision transformer, dubbed SalViT, for few-shot keypoint detection. Our SalViT enjoys a uniquely designed masked self-attention and a morphology learner, where the former introduces saliency map as a soft mask to constrain the self-attention on foregrounds, while the latter leverages the so-called power normalization to adjust morphology of saliency map, realizing ``dynamically changing receptive field''. Moreover, as salinecy detectors add computations, we show that attentive masks of DINO transformer can replace saliency. On top of SalViT, we also investigate i) transductive FSKD that enhances keypoint representations with unlabelled data and ii) FSKD under occlusions. We show that our model performs well on five public datasets and achieves ~10% PCK higher than the normally trained model under severe occlusions., Comment: 15 pages, 10 figures

Many skeletal action recognition models use GCNs to represent the human body by 3D body joints connected body parts. GCNs aggregate one- or few-hop graph neighbourhoods, and ignore the dependency between not linked body joints. We propose to form hypergraph to model hyper-edges between graph nodes (e.g., third- and fourth-order hyper-edges capture three and four nodes) which help capture higher-order motion patterns of groups of body joints. We split action sequences into temporal blocks, Higher-order Transformer (HoT) produces embeddings of each temporal block based on (i) the body joints, (ii) pairwise links of body joints and (iii) higher-order hyper-edges of skeleton body joints. We combine such HoT embeddings of hyper-edges of orders 1, ..., r by a novel Multi-order Multi-mode Transformer (3Mformer) with two modules whose order can be exchanged to achieve coupled-mode attention on coupled-mode tokens based on 'channel-temporal block', 'order-channel-body joint', 'channel-hyper-edge (any order)' and 'channel-only' pairs. The first module, called Multi-order Pooling (MP), additionally learns weighted aggregation along the hyper-edge mode, whereas the second module, Temporal block Pooling (TP), aggregates along the temporal block mode. Our end-to-end trainable network yields state-of-the-art results compared to GCN-, transformer- and hypergraph-based counterparts., Comment: This paper is accepted by CVPR 2023

Traffic forecasting is a critical task to extract values from cyber-physical infrastructures, which is the backbone of smart transportation. However owing to external contexts, the dynamics at each sensor are unique. For example, the afternoon peaks at sensors near schools are more likely to occur earlier than those near residential areas. In this paper, we first analyze real-world traffic data to show that each sensor has a unique dynamic. Further analysis also shows that each pair of sensors also has a unique dynamic. Then, we explore how node embedding learns the unique dynamics at every sensor location. Next, we propose a novel module called Spatial Graph Transformers (SGT) where we use node embedding to leverage the self-attention mechanism to ensure that the information flow between two sensors is adaptive with respect to the unique dynamic of each pair. Finally, we present Graph Self-attention WaveNet (G-SWaN) to address the complex, non-linear spatiotemporal traffic dynamics. Through empirical experiments on four real-world, open datasets, we show that the proposed method achieves superior performance on both traffic speed and flow forecasting. Code is available at: https://github.com/aprbw/G-SWaN, Comment: 20 pages, IoTDI 2023; Correction on Fig. 4

Contemporary deep learning multi-scale deblurring models suffer from many issues: 1) They perform poorly on non-uniformly blurred images/videos; 2) Simply increasing the model depth with finer-scale levels cannot improve deblurring; 3) Individual RGB frames contain a limited motion information for deblurring; 4) Previous models have a limited robustness to spatial transformations and noise. Below, we extend the DMPHN model by several mechanisms to address the above issues: I) We present a novel self-supervised event-guided deep hierarchical Multi-patch Network (MPN) to deal with blurry images and videos via fine-to-coarse hierarchical localized representations; II) We propose a novel stacked pipeline, StackMPN, to improve the deblurring performance under the increased network depth; III) We propose an event-guided architecture to exploit motion cues contained in videos to tackle complex blur in videos; IV) We propose a novel self-supervised step to expose the model to random transformations (rotations, scale changes), and make it robust to Gaussian noises. Our MPN achieves the state of the art on the GoPro and VideoDeblur datasets with a 40x faster runtime compared to current multi-scale methods. With 30ms to process an image at 1280x720 resolution, it is the first real-time deep motion deblurring model for 720p images at 30fps. For StackMPN, we obtain significant improvements over 1.2dB on the GoPro dataset by increasing the network depth. Utilizing the event information and self-supervision further boost results to 33.83dB., Comment: International Journal of Computer Vision. arXiv admin note: substantial text overlap with arXiv:1904.03468

Although augmentations (e.g., perturbation of graph edges, image crops) boost the efficiency of Contrastive Learning (CL), feature level augmentation is another plausible, complementary yet not well researched strategy. Thus, we present a novel spectral feature argumentation for contrastive learning on graphs (and images). To this end, for each data view, we estimate a low-rank approximation per feature map and subtract that approximation from the map to obtain its complement. This is achieved by the proposed herein incomplete power iteration, a non-standard power iteration regime which enjoys two valuable byproducts (under mere one or two iterations): (i) it partially balances spectrum of the feature map, and (ii) it injects the noise into rebalanced singular values of the feature map (spectral augmentation). For two views, we align these rebalanced feature maps as such an improved alignment step can focus more on less dominant singular values of matrices of both views, whereas the spectral augmentation does not affect the spectral angle alignment (singular vectors are not perturbed). We derive the analytical form for: (i) the incomplete power iteration to capture its spectrum-balancing effect, and (ii) the variance of singular values augmented implicitly by the noise. We also show that the spectral augmentation improves the generalization bound. Experiments on graph/image datasets show that our spectral feature augmentation outperforms baselines, and is complementary with other augmentation strategies and compatible with various contrastive losses., Comment: This paper has been published with the Thirty-Seventh AAAI Conference on Artificial Intelligence (AAAI 2023)

In this paper, we tackle the challenging problem of Few-shot Object Detection. Existing FSOD pipelines (i) use average-pooled representations that result in information loss; and/or (ii) discard position information that can help detect object instances. Consequently, such pipelines are sensitive to large intra-class appearance and geometric variations between support and query images. To address these drawbacks, we propose a Time-rEversed diffusioN tEnsor Transformer (TENET), which i) forms high-order tensor representations that capture multi-way feature occurrences that are highly discriminative, and ii) uses a transformer that dynamically extracts correlations between the query image and the entire support set, instead of a single average-pooled support embedding. We also propose a Transformer Relation Head (TRH), equipped with higher-order representations, which encodes correlations between query regions and the entire support set, while being sensitive to the positional variability of object instances. Our model achieves state-of-the-art results on PASCAL VOC, FSOD, and COCO., Comment: Accepted at the 17th European Conference on Computer Vision (ECCV 2022)

We propose a Few-shot Learning pipeline for 3D skeleton-based action recognition by Joint tEmporal and cAmera viewpoiNt alIgnmEnt (JEANIE). To factor out misalignment between query and support sequences of 3D body joints, we propose an advanced variant of Dynamic Time Warping which jointly models each smooth path between the query and support frames to achieve simultaneously the best alignment in the temporal and simulated camera viewpoint spaces for end-to-end learning under the limited few-shot training data. Sequences are encoded with a temporal block encoder based on Simple Spectral Graph Convolution, a lightweight linear Graph Neural Network backbone. We also include a setting with a transformer. Finally, we propose a similarity-based loss which encourages the alignment of sequences of the same class while preventing the alignment of unrelated sequences. We show state-of-the-art results on NTU-60, NTU-120, Kinetics-skeleton and UWA3D Multiview Activity II., Comment: Accepted as an oral paper at the 16th Asian Conference on Computer Vision (ACCV 2022). It extends our arXiv preprint arXiv:2112.12668 (2021)

Dynamic Time Warping (DTW) is used for matching pairs of sequences and celebrated in applications such as forecasting the evolution of time series, clustering time series or even matching sequence pairs in few-shot action recognition. The transportation plan of DTW contains a set of paths; each path matches frames between two sequences under a varying degree of time warping, to account for varying temporal intra-class dynamics of actions. However, as DTW is the smallest distance among all paths, it may be affected by the feature uncertainty which varies across time steps/frames. Thus, in this paper, we propose to model the so-called aleatoric uncertainty of a differentiable (soft) version of DTW. To this end, we model the heteroscedastic aleatoric uncertainty of each path by the product of likelihoods from Normal distributions, each capturing variance of pair of frames. (The path distance is the sum of base distances between features of pairs of frames of the path.) The Maximum Likelihood Estimation (MLE) applied to a path yields two terms: (i) a sum of Euclidean distances weighted by the variance inverse, and (ii) a sum of log-variance regularization terms. Thus, our uncertainty-DTW is the smallest weighted path distance among all paths, and the regularization term (penalty for the high uncertainty) is the aggregate of log-variances along the path. The distance and the regularization term can be used in various objectives. We showcase forecasting the evolution of time series, estimating the Fr\'echet mean of time series, and supervised/unsupervised few-shot action recognition of the articulated human 3D body joints., Comment: Accepted as an oral paper at the 17th European Conference on Computer Vision (ECCV 2022). arXiv admin note: text overlap with arXiv:2210.16820

Graph contrastive learning (GCL) improves graph representation learning, leading to SOTA on various downstream tasks. The graph augmentation step is a vital but scarcely studied step of GCL. In this paper, we show that the node embedding obtained via the graph augmentations is highly biased, somewhat limiting contrastive models from learning discriminative features for downstream tasks. Thus, instead of investigating graph augmentation in the input space, we alternatively propose to perform augmentations on the hidden features (feature augmentation). Inspired by so-called matrix sketching, we propose COSTA, a novel COvariance-preServing feaTure space Augmentation framework for GCL, which generates augmented features by maintaining a "good sketch" of original features. To highlight the superiority of feature augmentation with COSTA, we investigate a single-view setting (in addition to multi-view one) which conserves memory and computations. We show that the feature augmentation with COSTA achieves comparable/better results than graph augmentation based models., Comment: This paper is accepted by the ACM KDD 2022

We address the problem of ground-to-satellite image geo-localization, that is, estimating the camera latitude, longitude and orientation (azimuth angle) by matching a query image captured at the ground level against a large-scale database with geotagged satellite images. Our prior arts treat the above task as pure image retrieval by selecting the most similar satellite reference image matching the ground-level query image. However, such an approach often produces coarse location estimates because the geotag of the retrieved satellite image only corresponds to the image center while the ground camera can be located at any point within the image. To further consolidate our prior research findings, we present a novel geometry-aware geo-localization method. Our new method is able to achieve the fine-grained location of a query image, up to pixel size precision of the satellite image, once its coarse location and orientation have been determined. Moreover, we propose a new geometry-aware image retrieval pipeline to improve the coarse localization accuracy. Apart from a polar transform in our conference work, this new pipeline also maps satellite image pixels to the ground-level plane in the ground-view via a geometry-constrained projective transform to emphasize informative regions, such as road structures, for cross-view geo-localization. Extensive quantitative and qualitative experiments demonstrate the effectiveness of our newly proposed framework. We also significantly improve the performance of coarse localization results compared to the state-of-the-art in terms of location recalls., Comment: submitted to TPAMI in Jan 2021

Graph Neural Networks (GNNs) have achieved remarkable success by extending traditional convolution to learning on non-Euclidean data. The key to the GNNs is adopting the neural message-passing paradigm with two stages: aggregation and update. The current design of GNNs considers the topology information in the aggregation stage. However, in the updating stage, all nodes share the same updating function. The identical updating function treats each node embedding as i.i.d. random variables and thus ignores the implicit relationships between neighborhoods, which limits the capacity of the GNNs. The updating function is usually implemented with a linear transformation followed by a non-linear activation function. To make the updating function topology-aware, we inject the topological information into the non-linear activation function and propose Graph-adaptive Rectified Linear Unit (GReLU), which is a new parametric activation function incorporating the neighborhood information in a novel and efficient way. The parameters of GReLU are obtained from a hyperfunction based on both node features and the corresponding adjacent matrix. To reduce the risk of overfitting and the computational cost, we decompose the hyperfunction as two independent components for nodes and features respectively. We conduct comprehensive experiments to show that our plug-and-play GReLU method is efficient and effective given different GNN backbones and various downstream tasks., Comment: TheWebConf (WWW), 2022

We propose a Multi-level Second-order (MlSo) few-shot learning network for supervised or unsupervised few-shot image classification and few-shot action recognition. We leverage so-called power-normalized second-order base learner streams combined with features that express multiple levels of visual abstraction, and we use self-supervised discriminating mechanisms. As Second-order Pooling (SoP) is popular in image recognition, we employ its basic element-wise variant in our pipeline. The goal of multi-level feature design is to extract feature representations at different layer-wise levels of CNN, realizing several levels of visual abstraction to achieve robust few-shot learning. As SoP can handle convolutional feature maps of varying spatial sizes, we also introduce image inputs at multiple spatial scales into MlSo. To exploit the discriminative information from multi-level and multi-scale features, we develop a Feature Matching (FM) module that reweights their respective branches. We also introduce a self-supervised step, which is a discriminator of the spatial level and the scale of abstraction. Our pipeline is trained in an end-to-end manner. With a simple architecture, we demonstrate respectable results on standard datasets such as Omniglot, mini-ImageNet, tiered-ImageNet, Open MIC, fine-grained datasets such as CUB Birds, Stanford Dogs and Cars, and action recognition datasets such as HMDB51, UCF101, and mini-MIT., Comment: IEEE Transactions on Multimedia

Graph contrastive learning attracts/disperses node representations for similar/dissimilar node pairs under some notion of similarity. It may be combined with a low-dimensional embedding of nodes to preserve intrinsic and structural properties of a graph. In this paper, we extend the celebrated Laplacian Eigenmaps with contrastive learning, and call them COntrastive Laplacian EigenmapS (COLES). Starting from a GAN-inspired contrastive formulation, we show that the Jensen-Shannon divergence underlying many contrastive graph embedding models fails under disjoint positive and negative distributions, which may naturally emerge during sampling in the contrastive setting. In contrast, we demonstrate analytically that COLES essentially minimizes a surrogate of Wasserstein distance, which is known to cope well under disjoint distributions. Moreover, we show that the loss of COLES belongs to the family of so-called block-contrastive losses, previously shown to be superior compared to pair-wise losses typically used by contrastive methods. We show on popular benchmarks/backbones that COLES offers favourable accuracy/scalability compared to DeepWalk, GCN, Graph2Gauss, DGI and GRACE baselines., Comment: Accepted by NeurIPS 2021. Includes the main paper and the supplementary material. OpenReview: https://openreview.net/forum?id=iLn-bhP-kKH

In this paper, we propose a Few-shot Learning pipeline for 3D skeleton-based action recognition by Joint tEmporal and cAmera viewpoiNt alIgnmEnt (JEANIE). To factor out misalignment between query and support sequences of 3D body joints, we propose an advanced variant of Dynamic Time Warping which jointly models each smooth path between the query and support frames to achieve simultaneously the best alignment in the temporal and simulated camera viewpoint spaces for end-to-end learning under the limited few-shot training data. Sequences are encoded with a temporal block encoder based on Simple Spectral Graph Convolution, a lightweight linear Graph Neural Network backbone (we also include a setting with a transformer). Finally, we propose a similarity-based loss which encourages the alignment of sequences of the same class while preventing the alignment of unrelated sequences. We demonstrate state-of-the-art results on NTU-60, NTU-120, Kinetics-skeleton and UWA3D Multiview Activity II., Comment: Full 17 page version

In this paper, we improve Generative Adversarial Networks by incorporating a manifold learning step into the discriminator. We consider locality-constrained linear and subspace-based manifolds, and locality-constrained non-linear manifolds. In our design, the manifold learning and coding steps are intertwined with layers of the discriminator, with the goal of attracting intermediate feature representations onto manifolds. We adaptively balance the discrepancy between feature representations and their manifold view, which is a trade-off between denoising on the manifold and refining the manifold. We find that locality-constrained non-linear manifolds outperform linear manifolds due to their non-uniform density and smoothness. We also substantially outperform state-of-the-art baselines., Comment: CVPR 2022, 32 pages full version

Current non-rigid object keypoint detectors perform well on a chosen kind of species and body parts, and require a large amount of labelled keypoints for training. Moreover, their heatmaps, tailored to specific body parts, cannot recognize novel keypoints (keypoints not labelled for training) on unseen species. We raise an interesting yet challenging question: how to detect both base (annotated for training) and novel keypoints for unseen species given a few annotated samples? Thus, we propose a versatile Few-shot Keypoint Detection (FSKD) pipeline, which can detect a varying number of keypoints of different kinds. Our FSKD provides the uncertainty estimation of predicted keypoints. Specifically, FSKD involves main and auxiliary keypoint representation learning, similarity learning, and keypoint localization with uncertainty modeling to tackle the localization noise. Moreover, we model the uncertainty across groups of keypoints by multivariate Gaussian distribution to exploit implicit correlations between neighboring keypoints. We show the effectiveness of our FSKD on (i) novel keypoint detection for unseen species, and (ii) few-shot Fine-Grained Visual Recognition (FGVR) and (iii) Semantic Alignment (SA) downstream tasks. For FGVR, detected keypoints improve the classification accuracy. For SA, we showcase a novel thin-plate-spline warping that uses estimated keypoint uncertainty under imperfect keypoint corespondences., Comment: Accepted by CVPR 2022; 8 pages for main paper, 6 pages for supplementary materials

Even with the luxury of having abundant data, multi-label classification is widely known to be a challenging task to address. This work targets the problem of multi-label meta-learning, where a model learns to predict multiple labels within a query (e.g., an image) by just observing a few supporting examples. In doing so, we first propose a benchmark for Few-Shot Learning (FSL) with multiple labels per sample. Next, we discuss and extend several solutions specifically designed to address the conventional and single-label FSL, to work in the multi-label regime. Lastly, we introduce a neural module to estimate the label count of a given sample by exploiting the relational inference. We will show empirically the benefit of the label count module, the label propagation algorithm, and the extensions of conventional FSL methods on three challenging datasets, namely MS-COCO, iMaterialist, and Open MIC. Overall, our thorough experiments suggest that the proposed label-propagation algorithm in conjunction with the neural label count module (NLC) shall be considered as the method of choice., Comment: Accepted to WACV 2022

Neural Architecture Search (NAS) is the game changer in designing robust neural architectures. Architectures designed by NAS outperform or compete with the best manual network designs in terms of accuracy, size, memory footprint and FLOPs. That said, previous studies focus on developing NAS algorithms for clean high quality data, a restrictive and somewhat unrealistic assumption. In this paper, focusing on the differentiable NAS algorithms, we show that vanilla NAS algorithms suffer from a performance loss if class labels are noisy. To combat this issue, we make use of the principle of information bottleneck as a regularizer. This leads us to develop a noise injecting operation that is included during the learning process, preventing the network from learning from noisy samples. Our empirical evaluations show that the noise injecting operation does not degrade the performance of the NAS algorithm if the data is indeed clean. In contrast, if the data is noisy, the architecture learned by our algorithm comfortably outperforms algorithms specifically equipped with sophisticated mechanisms to learn in the presence of label noise. In contrast to many algorithms designed to work in the presence of noisy labels, prior knowledge about the properties of the noise and its characteristics are not required for our algorithm., Comment: Accepted to WACV 2022

We devise a multimodal conversation system for dialogue utterances composed of text, image or both modalities. We leverage Auxiliary UnsuperviseD vIsual and TExtual Data (AUDITED). To improve the performance of text-based task, we utilize translations of target sentences from English to French to form the assisted supervision. For the image-based task, we employ the DeepFashion dataset in which we seek nearest neighbor images of positive and negative target images of the MMD data. These nearest neighbors form the nearest neighbor embedding providing an external context for target images. We form two methods to create neighbor embedding vectors, namely Neighbor Embedding by Hard Assignment (NEHA) and Neighbor Embedding by Soft Assignment (NESA) which generate context subspaces per target image. Subsequently, these subspaces are learnt by our pipeline as a context for the target data. We also propose a discriminator which switches between the image- and text-based tasks. We show improvements over baselines on the large-scale Multimodal Dialogue Dataset (MMD) and SIMMC., Comment: Accepted by the BMVC 2021

We aim at capturing high-order statistics of feature vectors formed by a neural network, and propose end-to-end second- and higher-order pooling to form a tensor descriptor. Tensor descriptors require a robust similarity measure due to low numbers of aggregated vectors and the burstiness phenomenon, when a given feature appears more/less frequently than statistically expected. The Heat Diffusion Process (HDP) on a graph Laplacian is closely related to the Eigenvalue Power Normalization (EPN) of the covariance/autocorrelation matrix, whose inverse forms a loopy graph Laplacian. We show that the HDP and the EPN play the same role, i.e., to boost or dampen the magnitude of the eigenspectrum thus preventing the burstiness. We equip higher-order tensors with EPN which acts as a spectral detector of higher-order occurrences to prevent burstiness. We also prove that for a tensor of order r built from d dimensional feature descriptors, such a detector gives the likelihood if at least one higher-order occurrence is 'projected' into one of binom(d,r) subspaces represented by the tensor; thus forming a tensor power normalization metric endowed with binom(d,r) such 'detectors'. For experimental contributions, we apply several second- and higher-order pooling variants to action recognition, provide previously not presented comparisons of such pooling variants, and show state-of-the-art results on HMDB-51, YUP++ and MPII Cooking Activities., Comment: Accepted by IEEE International Conference on Acoustics, Speech, and Signal Processing (ICASSP 2024)

Network embedding approaches have recently attracted considerable interest as they learn low-dimensional vector representations of nodes. Embeddings based on the matrix factorization are effective but they are usually computationally expensive due to the eigen-decomposition step. In this paper, we propose a Random RangE FInder based Network Embedding (REFINE) algorithm, which can perform embedding on one million of nodes (YouTube) within 30 seconds in a single thread. REFINE is 10x faster than ProNE, which is 10-400x faster than other methods such as LINE, DeepWalk, Node2Vec, GraRep, and Hope. Firstly, we formulate our network embedding approach as a skip-gram model, but with an orthogonal constraint, and we reformulate it into the matrix factorization problem. Instead of using randomized tSVD (truncated SVD) as other methods, we employ the Randomized Blocked QR decomposition to obtain the node representation fast. Moreover, we design a simple but efficient spectral filter for network enhancement to obtain higher-order information for node representation. Experimental results prove that REFINE is very efficient on datasets of different sizes (from thousand to million of nodes/edges) for node classification, while enjoying a good performance.

Although Graph Convolutional Networks (GCNs) have demonstrated their power in various applications, the graph convolutional layers, as the most important component of GCN, are still using linear transformations and a simple pooling step. In this paper, we propose a novel generalization of Factorized Bilinear (FB) layer to model the feature interactions in GCNs. FB performs two matrix-vector multiplications, that is, the weight matrix is multiplied with the outer product of the vector of hidden features from both sides. However, the FB layer suffers from the quadratic number of coefficients, overfitting and the spurious correlations due to correlations between channels of hidden representations that violate the i.i.d. assumption. Thus, we propose a compact FB layer by defining a family of summarizing operators applied over the quadratic term. We analyze proposed pooling operators and motivate their use. Our experimental results on multiple datasets demonstrate that the GFB-GCN is competitive with other methods for text classification.

To model and forecast flight delays accurately, it is crucial to harness various vehicle trajectory and contextual sensor data on airport tarmac areas. These heterogeneous sensor data, if modelled correctly, can be used to generate a situational awareness map. Existing techniques apply traditional supervised learning methods onto historical data, contextual features and route information among different airports to predict flight delay are inaccurate and only predict arrival delay but not departure delay, which is essential to airlines. In this paper, we propose a vision-based solution to achieve a high forecasting accuracy, applicable to the airport. Our solution leverages a snapshot of the airport situational awareness map, which contains various trajectories of aircraft and contextual features such as weather and airline schedules. We propose an end-to-end deep learning architecture, TrajCNN, which captures both the spatial and temporal information from the situational awareness map. Additionally, we reveal that the situational awareness map of the airport has a vital impact on estimating flight departure delay. Our proposed framework obtained a good result (around 18 minutes error) for predicting flight departure delay at Los Angeles International Airport., Comment: single column. Neurocomputing, 2021

Neural networks notoriously suffer from the problem of catastrophic forgetting, the phenomenon of forgetting the past knowledge when acquiring new knowledge. Overcoming catastrophic forgetting is of significant importance to emulate the process of "incremental learning", where the model is capable of learning from sequential experience in an efficient and robust way. State-of-the-art techniques for incremental learning make use of knowledge distillation towards preventing catastrophic forgetting. Therein, one updates the network while ensuring that the network's responses to previously seen concepts remain stable throughout updates. This in practice is done by minimizing the dissimilarity between current and previous responses of the network one way or another. Our work contributes a novel method to the arsenal of distillation techniques. In contrast to the previous state of the art, we propose to firstly construct low-dimensional manifolds for previous and current responses and minimize the dissimilarity between the responses along the geodesic connecting the manifolds. This induces a more formidable knowledge distillation with smooth properties which preserves the past knowledge more efficiently as observed by our comprehensive empirical study., Comment: Accepted to CVPR 2021

Human actions in video sequences are characterized by the complex interplay between spatial features and their temporal dynamics. In this paper, we propose novel tensor representations for compactly capturing such higher-order relationships between visual features for the task of action recognition. We propose two tensor-based feature representations, viz. (i) sequence compatibility kernel (SCK) and (ii) dynamics compatibility kernel (DCK). SCK builds on the spatio-temporal correlations between features, whereas DCK explicitly models the action dynamics of a sequence. We also explore generalization of SCK, coined SCK(+), that operates on subsequences to capture the local-global interplay of correlations, which can incorporate multi-modal inputs e.g., skeleton 3D body-joints and per-frame classifier scores obtained from deep learning models trained on videos. We introduce linearization of these kernels that lead to compact and fast descriptors. We provide experiments on (i) 3D skeleton action sequences, (ii) fine-grained video sequences, and (iii) standard non-fine-grained videos. As our final representations are tensors that capture higher-order relationships of features, they relate to co-occurrences for robust fine-grained recognition. We use higher-order tensors and so-called Eigenvalue Power Normalization (EPN) which have been long speculated to perform spectral detection of higher-order occurrences, thus detecting fine-grained relationships of features rather than merely count features in action sequences. We prove that a tensor of order r, built from Z* dimensional features, coupled with EPN indeed detects if at least one higher-order occurrence is `projected' into one of its binom(Z*,r) subspaces of dim. r represented by the tensor, thus forming a Tensor Power Normalization metric endowed with binom(Z*,r) such `detectors'., Comment: Published with TPAMI, 2020. arXiv admin note: text overlap with arXiv:1604.00239

Power Normalizations (PN) are useful non-linear operators which tackle feature imbalances in classification problems. We study PNs in the deep learning setup via a novel PN layer pooling feature maps. Our layer combines the feature vectors and their respective spatial locations in the feature maps produced by the last convolutional layer of CNN into a positive definite matrix with second-order statistics to which PN operators are applied, forming so-called Second-order Pooling (SOP). As the main goal of this paper is to study Power Normalizations, we investigate the role and meaning of MaxExp and Gamma, two popular PN functions. To this end, we provide probabilistic interpretations of such element-wise operators and discover surrogates with well-behaved derivatives for end-to-end training. Furthermore, we look at the spectral applicability of MaxExp and Gamma by studying Spectral Power Normalizations (SPN). We show that SPN on the autocorrelation/covariance matrix and the Heat Diffusion Process (HDP) on a graph Laplacian matrix are closely related, thus sharing their properties. Such a finding leads us to the culmination of our work, a fast spectral MaxExp which is a variant of HDP for covariances/autocorrelation matrices. We evaluate our ideas on fine-grained recognition, scene recognition, and material classification, as well as in few-shot learning and graph classification., Comment: Accepted by TPAMI, July 2020

Sentence compression is a Natural Language Processing (NLP) task aimed at shortening original sentences and preserving their key information. Its applications can benefit many fields e.g. one can build tools for language education. However, current methods are largely based on Recurrent Neural Network (RNN) models which suffer from poor processing speed. To address this issue, in this paper, we propose a token-wise Convolutional Neural Network, a CNN-based model along with pre-trained Bidirectional Encoder Representations from Transformers (BERT) features for deletion-based sentence compression. We also compare our model with RNN-based models and fine-tuned BERT. Although one of the RNN-based models outperforms marginally other models given the same input, our CNN-based model was ten times faster than the RNN-based approach.

Estimating a 6DOF object pose from a single image is very challenging due to occlusions or textureless appearances. Vector-field based keypoint voting has demonstrated its effectiveness and superiority on tackling those issues. However, direct regression of vector-fields neglects that the distances between pixels and keypoints also affect the deviations of hypotheses dramatically. In other words, small errors in direction vectors may generate severely deviated hypotheses when pixels are far away from a keypoint. In this paper, we aim to reduce such errors by incorporating the distances between pixels and keypoints into our objective. To this end, we develop a simple yet effective differentiable proxy voting loss (DPVL) which mimics the hypothesis selection in the voting procedure. By exploiting our voting loss, we are able to train our network in an end-to-end manner. Experiments on widely used datasets, i.e., LINEMOD and Occlusion LINEMOD, manifest that our DPVL improves pose estimation performance significantly and speeds up the training convergence.

Point-of-Interest (POI) recommendation is one of the most important location-based services helping people discover interesting venues or services. However, the extreme user-POI matrix sparsity and the varying spatio-temporal context pose challenges for POI systems, which affects the quality of POI recommendations. To this end, we propose a translation-based relation embedding for POI recommendation. Our approach encodes the temporal and geographic information, as well as semantic contents effectively in a low-dimensional relation space by using Knowledge Graph Embedding techniques. To further alleviate the issue of user-POI matrix sparsity, a combined matrix factorization framework is built on a user-POI graph to enhance the inference of dynamic personal interests by exploiting the side-information. Experiments on two real-world datasets demonstrate the effectiveness of our proposed model., Comment: 12 pages, 3 figures, Accepted in the 24th Pacific-Asia Conference on Knowledge Discovery and Data Mining (PAKDD 2020)

In this paper, we build on a concept of self-supervision by taking RGB frames as input to learn to predict both action concepts and auxiliary descriptors e.g., object descriptors. So-called hallucination streams are trained to predict auxiliary cues, simultaneously fed into classification layers, and then hallucinated at the testing stage to aid network. We design and hallucinate two descriptors, one leveraging four popular object detectors applied to training videos, and the other leveraging image- and video-level saliency detectors. The first descriptor encodes the detector- and ImageNet-wise class prediction scores, confidence scores, and spatial locations of bounding boxes and frame indexes to capture the spatio-temporal distribution of features per video. Another descriptor encodes spatio-angular gradient distributions of saliency maps and intensity patterns. Inspired by the characteristic function of the probability distribution, we capture four statistical moments on the above intermediate descriptors. As numbers of coefficients in the mean, covariance, coskewness and cokurtotsis grow linearly, quadratically, cubically and quartically w.r.t. the dimension of feature vectors, we describe the covariance matrix by its leading n' eigenvectors (so-called subspace) and we capture skewness/kurtosis rather than costly coskewness/cokurtosis. We obtain state of the art on five popular datasets such as Charades and EPIC-Kitchens., Comment: ACM MM'21