Minh N. Do

Minh-Quan Le, Tam V. Nguyen, Trung-Nghia Le et al.

Few-shot instance segmentation extends the few-shot learning paradigm to the instance segmentation task, which tries to segment instance objects from a query image with a few annotated examples of novel categories. Conventional approaches have attempted to address the task via prototype learning, known as point estimation. However, this mechanism depends on prototypes (\eg mean of $K-$shot) for prediction, leading to performance instability. To overcome the disadvantage of the point estimation mechanism, we propose a novel approach, dubbed MaskDiff, which models the underlying conditional distribution of a binary mask, which is conditioned on an object region and $K-$shot information. Inspired by augmentation approaches that perturb data with Gaussian noise for populating low data density regions, we model the mask distribution with a diffusion probabilistic model. We also propose to utilize classifier-free guided mask sampling to integrate category information into the binary mask generation process. Without bells and whistles, our proposed method consistently outperforms state-of-the-art methods on both base and novel classes of the COCO dataset while simultaneously being more stable than existing methods. The source code is available at: https://github.com/minhquanlecs/MaskDiff.

Johann A. Bengua, Ho N. Phien, Hoang D. Tuan et al.

This paper proposes a novel approach to tensor completion, which recovers missing entries of data represented by tensors. The approach is based on the tensor train (TT) rank, which is able to capture hidden information from tensors thanks to its definition from a well-balanced matricization scheme. Accordingly, new optimization formulations for tensor completion are proposed as well as two new algorithms for their solution. The first one called simple low-rank tensor completion via tensor train (SiLRTC-TT) is intimately related to minimizing a nuclear norm based on TT rank. The second one is from a multilinear matrix factorization model to approximate the TT rank of a tensor, and is called tensor completion by parallel matrix factorization via tensor train (TMac-TT). A tensor augmentation scheme of transforming a low-order tensor to higher-orders is also proposed to enhance the effectiveness of SiLRTC-TT and TMac-TT. Simulation results for color image and video recovery show the clear advantage of our method over all other methods.

Khoi-Nguyen C. Mac, Minh N. Do, Minh P. Vo

Adaptive sampling that exploits the spatiotemporal redundancy in videos is critical for always-on action recognition on wearable devices with limited computing and battery resources. The commonly used fixed sampling strategy is not context-aware and may under-sample the visual content, and thus adversely impacts both computation efficiency and accuracy. Inspired by the concepts of foveal vision and pre-attentive processing from the human visual perception mechanism, we introduce a novel adaptive spatiotemporal sampling scheme for efficient action recognition. Our system pre-scans the global scene context at low-resolution and decides to skip or request high-resolution features at salient regions for further processing. We validate the system on EPIC-KITCHENS and UCF-101 datasets for action recognition, and show that our proposed approach can greatly speed up inference with a tolerable loss of accuracy compared with those from state-of-the-art baselines. Source code is available in https://github.com/knmac/adaptive_spatiotemporal.

Quan Nguyen, Hieu H. Pham, Kok-Seng Wong et al. · cmu, deepmind

We introduce FedDCT, a novel distributed learning paradigm that enables the usage of large, high-performance CNNs on resource-limited edge devices. As opposed to traditional FL approaches, which require each client to train the full-size neural network independently during each training round, the proposed FedDCT allows a cluster of several clients to collaboratively train a large deep learning model by dividing it into an ensemble of several small sub-models and train them on multiple devices in parallel while maintaining privacy. In this collaborative training process, clients from the same cluster can also learn from each other, further improving their ensemble performance. In the aggregation stage, the server takes a weighted average of all the ensemble models trained by all the clusters. FedDCT reduces the memory requirements and allows low-end devices to participate in FL. We empirically conduct extensive experiments on standardized datasets, including CIFAR-10, CIFAR-100, and two real-world medical datasets HAM10000 and VAIPE. Experimental results show that FedDCT outperforms a set of current SOTA FL methods with interesting convergence behaviors. Furthermore, compared to other existing approaches, FedDCT achieves higher accuracy and substantially reduces the number of communication rounds (with $4-8$ times fewer memory requirements) to achieve the desired accuracy on the testing dataset without incurring any extra training cost on the server side.

Renan A. Rojas-Gomez, Teck-Yian Lim, Alexander G. Schwing et al.

We propose learnable polyphase sampling (LPS), a pair of learnable down/upsampling layers that enable truly shift-invariant and equivariant convolutional networks. LPS can be trained end-to-end from data and generalizes existing handcrafted downsampling layers. It is widely applicable as it can be integrated into any convolutional network by replacing down/upsampling layers. We evaluate LPS on image classification and semantic segmentation. Experiments show that LPS is on-par with or outperforms existing methods in both performance and shift consistency. For the first time, we achieve true shift-equivariance on semantic segmentation (PASCAL VOC), i.e., 100% shift consistency, outperforming baselines by an absolute 3.3%.

Ho N. Phien, Hoang D. Tuan, Johann A. Bengua et al.

This paper proposes a novel formulation of the tensor completion problem to impute missing entries of data represented by tensors. The formulation is introduced in terms of tensor train (TT) rank which can effectively capture global information of tensors thanks to its construction by a well-balanced matricization scheme. Two algorithms are proposed to solve the corresponding tensor completion problem. The first one called simple low-rank tensor completion via tensor train (SiLRTC-TT) is intimately related to minimizing the TT nuclear norm. The second one is based on a multilinear matrix factorization model to approximate the TT rank of the tensor and called tensor completion by parallel matrix factorization via tensor train (TMac-TT). These algorithms are applied to complete both synthetic and real world data tensors. Simulation results of synthetic data show that the proposed algorithms are efficient in estimating missing entries for tensors with either low Tucker rank or TT rank while Tucker-based algorithms are only comparable in the case of low Tucker rank tensors. When applied to recover color images represented by ninth-order tensors augmented from third-order ones, the proposed algorithms outperforms the Tucker-based algorithms.

Trung-Hieu Hoang, Mona Zehni, Huaijin Xu et al.

The ability to use digitally recorded and quantified neurological exam information is important to help healthcare systems deliver better care, in-person and via telehealth, as they compensate for a growing shortage of neurologists. Current neurological digital biomarker pipelines, however, are narrowed down to a specific neurological exam component or applied for assessing specific conditions. In this paper, we propose an accessible vision-based exam and documentation solution called Digitized Neurological Examination (DNE) to expand exam biomarker recording options and clinical applications using a smartphone/tablet. Through our DNE software, healthcare providers in clinical settings and people at home are enabled to video capture an examination while performing instructed neurological tests, including finger tapping, finger to finger, forearm roll, and stand-up and walk. Our modular design of the DNE software supports integrations of additional tests. The DNE extracts from the recorded examinations the 2D/3D human-body pose and quantifies kinematic and spatio-temporal features. The features are clinically relevant and allow clinicians to document and observe the quantified movements and the changes of these metrics over time. A web server and a user interface for recordings viewing and feature visualizations are available. DNE was evaluated on a collected dataset of 21 subjects containing normal and simulated-impaired movements. The overall accuracy of DNE is demonstrated by classifying the recorded movements using various machine learning models. Our tests show an accuracy beyond 90% for upper-limb tests and 80% for the stand-up and walk tests.

Trung-Hieu Hoang, Duc Minh Vo, Minh N. Do

Current test-time adaptation (TTA) approaches aim to adapt a machine learning model to environments that change continuously. Yet, it is unclear whether TTA methods can maintain their adaptability over prolonged periods. To answer this question, we introduce a diagnostic setting - recurring TTA where environments not only change but also recur over time, creating an extensive data stream. This setting allows us to examine the error accumulation of TTA models, in the most basic scenario, when they are regularly exposed to previous testing environments. Furthermore, we simulate a TTA process on a simple yet representative $ε$-perturbed Gaussian Mixture Model Classifier, deriving theoretical insights into the dataset- and algorithm-dependent factors contributing to gradual performance degradation. Our investigation leads us to propose persistent TTA (PeTTA), which senses when the model is diverging towards collapse and adjusts the adaptation strategy, striking a balance between the dual objectives of adaptation and model collapse prevention. The supreme stability of PeTTA over existing approaches, in the face of lifelong TTA scenarios, has been demonstrated over comprehensive experiments on various benchmarks. Our project page is available at https://hthieu166.github.io/petta.

Abhi Kamboj, Anh Duy Nguyen, Minh N. Do

In order to unlock the potential of diverse sensors, we investigate a method to transfer knowledge between time-series modalities using a multimodal \textit{temporal} representation space for Human Activity Recognition (HAR). Specifically, we explore the setting where the modality used in testing has no labeled data during training, which we refer to as Unsupervised Modality Adaptation (UMA). We categorize existing UMA approaches as Student-Teacher or Contrastive Alignment methods. These methods typically compress continuous-time data samples into single latent vectors during alignment, inhibiting their ability to transfer temporal information through real-world temporal distortions. To address this, we introduce Cross-modal Transfer Through Time (C3T), which preserves temporal information during alignment to handle dynamic sensor data better. C3T achieves this by aligning a set of temporal latent vectors across sensing modalities. Our extensive experiments on various camera+IMU datasets demonstrate that C3T outperforms existing methods in UMA by at least 8% in accuracy and shows superior robustness to temporal distortions such as time-shift, misalignment, and dilation. Our findings suggest that C3T has significant potential for developing generalizable models for time-series sensor data, opening new avenues for various multimodal applications.

Duc N. Do, Minh N. Do, Dang Nguyen et al.

Transthoracic echocardiography is the reference standard for confirming structural heart disease (SHD), but first-line screening is limited by cost, workflow burden, and specialist availability. We evaluated whether open pretrained electrocardiogram (ECG) foundation models can support echo-confirmed multi-label SHD detection using the public EchoNext Mini-Model benchmark. Six echocardiography-derived abnormalities were targeted: reduced left ventricular ejection fraction, increased left ventricular wall thickness, aortic stenosis, mitral regurgitation, tricuspid regurgitation, and right ventricular systolic dysfunction. Under a common pipeline, we compared engineered ECG features with gradient boosting, end-to-end waveform learning from scratch, and transfer from open ECG foundation models. We then applied in-domain self-supervised adaptation of an ECG foundation model (ECG-FM) on EchoNext waveforms followed by selective supervised fine-tuning, and evaluated trade-offs between discrimination and adaptation cost. Adapted ECG-FM models achieved the best overall performance: peak macro-AUROC 0.8509 and macro-AUPRC 0.4297, while a parameter-efficient operating point preserved AUROC (0.8501) and attained the highest fixed-threshold macro-F1 0.3691. Late fusion with covariates did not improve threshold-independent discrimination, and evaluated LoRA, alternative backbones, and mixture-of-foundations strategies did not surpass the best adapted single-backbone models. These results indicate that for ECG-based case finding and echocardiography triage, combining target-domain self-supervised adaptation with selective supervised updating of a pretrained ECG backbone is the most effective transfer strategy.

Trung-Hieu Hoang, Mona Zehni, Huy Phan et al.

Despite the promising performance of current 3D human pose estimation techniques, understanding and enhancing their generalization on challenging in-the-wild videos remain an open problem. In this work, we focus on the robustness of 2D-to-3D pose lifters. To this end, we develop two benchmark datasets, namely Human3.6M-C and HumanEva-I-C, to examine the robustness of video-based 3D pose lifters to a wide range of common video corruptions including temporary occlusion, motion blur, and pixel-level noise. We observe the poor generalization of state-of-the-art 3D pose lifters in the presence of corruption and establish two techniques to tackle this issue. First, we introduce Temporal Additive Gaussian Noise (TAGN) as a simple yet effective 2D input pose data augmentation. Additionally, to incorporate the confidence scores output by the 2D pose detectors, we design a confidence-aware convolution (CA-Conv) block. Extensively tested on corrupted videos, the proposed strategies consistently boost the robustness of 3D pose lifters and serve as new baselines for future research.

Renan A. Rojas-Gomez, Minh N. Do

State-of-the-art Style Transfer methods often leverage pre-trained encoders optimized for discriminative tasks, which may not be ideal for image synthesis. This can result in significant artifacts and loss of photorealism. Motivated by the ability of multiscale geometric image representations to capture fine-grained details and global structure, we propose GIST: Geometric-based Image Style Transfer, a novel Style Transfer technique that exploits the geometric properties of content and style images. GIST replaces the standard Neural Style Transfer autoencoding framework with a multiscale image expansion, preserving scene details without the need for post-processing or training. Our method matches multiresolution and multidirectional representations such as Wavelets and Contourlets by solving an optimal transport problem, leading to an efficient texture transferring. Experiments show that GIST is on-par or outperforms recent photorealistic Style Transfer approaches while significantly reducing the processing time with no model training.

Vida Adeli, Ivan Klabucar, Javad Rajabi et al.

Objective gait assessment in Parkinson's Disease (PD) is limited by the absence of large, diverse, and clinically annotated motion datasets. We introduce CARE-PD, the largest publicly available archive of 3D mesh gait data for PD, and the first multi-site collection spanning 9 cohorts from 8 clinical centers. All recordings (RGB video or motion capture) are converted into anonymized SMPL meshes via a harmonized preprocessing pipeline. CARE-PD supports two key benchmarks: supervised clinical score prediction (estimating Unified Parkinson's Disease Rating Scale, UPDRS, gait scores) and unsupervised motion pretext tasks (2D-to-3D keypoint lifting and full-body 3D reconstruction). Clinical prediction is evaluated under four generalization protocols: within-dataset, cross-dataset, leave-one-dataset-out, and multi-dataset in-domain adaptation. To assess clinical relevance, we compare state-of-the-art motion encoders with a traditional gait-feature baseline, finding that encoders consistently outperform handcrafted features. Pretraining on CARE-PD reduces MPJPE (from 60.8mm to 7.5mm) and boosts PD severity macro-F1 by 17 percentage points, underscoring the value of clinically curated, diverse training data. CARE-PD and all benchmark code are released for non-commercial research at https://neurips2025.care-pd.ca/.

Duy A. Nguyen, Abhi Kamboj, Minh N. Do

Addressing missing modalities and limited labeled data is crucial for advancing robust multimodal learning. We propose Robult, a scalable framework designed to mitigate these challenges by preserving modality-specific information and leveraging redundancy through a novel information-theoretic approach. Robult optimizes two core objectives: (1) a soft Positive-Unlabeled (PU) contrastive loss that maximizes task-relevant feature alignment while effectively utilizing limited labeled data in semi-supervised settings, and (2) a latent reconstruction loss that ensures unique modality-specific information is retained. These strategies, embedded within a modular design, enhance performance across various downstream tasks and ensure resilience to incomplete modalities during inference. Experimental results across diverse datasets validate that Robult achieves superior performance over existing approaches in both semi-supervised learning and missing modality contexts. Furthermore, its lightweight design promotes scalability and seamless integration with existing architectures, making it suitable for real-world multimodal applications.

Duy A. Nguyen, Quan Huu Do, Khoa D. Doan et al.

Multimodal learning has demonstrated incredible successes by integrating diverse data sources, yet it often relies on the availability of all modalities - an assumption that rarely holds in real-world applications. Pretrained multimodal models, while effective, struggle when confronted with small-scale and incomplete datasets (i.e., missing modalities), limiting their practical applicability. Previous studies on reconstructing missing modalities have overlooked the reconstruction's potential unreliability, which could compromise the quality of the final outputs. We present SURE (Scalable Uncertainty and Reconstruction Estimation), a novel framework that extends the capabilities of pretrained multimodal models by introducing latent space reconstruction and uncertainty estimation for both reconstructed modalities and downstream tasks. Our method is architecture-agnostic, reconstructs missing modalities, and delivers reliable uncertainty estimates, improving both interpretability and performance. SURE introduces a unique Pearson Correlation-based loss and applies statistical error propagation in deep networks for the first time, allowing precise quantification of uncertainties from missing data and model predictions. Extensive experiments across tasks such as sentiment analysis, genre classification, and action recognition show that SURE consistently achieves state-of-the-art performance, ensuring robust predictions even in the presence of incomplete data.

Abhi Kamboj, Minh N. Do

Multimodal alignment constructs a joint latent vector space where modalities representing the same concept map to neighboring latent vectors. We formulate this as an inverse problem and show that, under certain conditions, paired data from each modality can map to equivalent latent vectors, which we refer to as perfect alignment. When perfect alignment cannot be achieved, it can be approximated using the Singular Value Decomposition (SVD) of a multimodal data matrix. Experiments on synthetic multimodal Gaussian data verify the effectiveness of our perfect alignment method compared to a learned contrastive alignment method. We further demonstrate the practical application of cross-modal transfer for human action recognition, showing that perfect alignment significantly enhances the model's accuracy. We conclude by discussing how these findings can be applied to various modalities and tasks and the limitations of our method. We hope these findings inspire further exploration of perfect alignment and its applications in representation learning.

Trung-Hieu Hoang, Duc Minh Vo, Minh N. Do

Test-time adaptation (TTA) has emerged as a promising solution to tackle the continual domain shift in machine learning by allowing model parameters to change at test time, via self-supervised learning on unlabeled testing data. At the same time, it unfortunately opens the door to unforeseen vulnerabilities for degradation over time. Through a simple theoretical continual TTA model, we successfully identify a risk in the sampling process of testing data that could easily degrade the performance of a continual TTA model. We name this risk as Reusing of Incorrect Prediction (RIP) that TTA attackers can employ or as a result of the unintended query from general TTA users. The risk posed by RIP is also highly realistic, as it does not require prior knowledge of model parameters or modification of testing samples. This simple requirement makes RIP as the first black-box TTA attack algorithm that stands out from existing white-box attempts. We extensively benchmark the performance of the most recent continual TTA approaches when facing the RIP attack, providing insights on its success, and laying out potential roadmaps that could enhance the resilience of future continual TTA systems.

Renan A. Rojas-Gomez, Teck-Yian Lim, Minh N. Do et al.

For computer vision, Vision Transformers (ViTs) have become one of the go-to deep net architectures. Despite being inspired by Convolutional Neural Networks (CNNs), ViTs' output remains sensitive to small spatial shifts in the input, i.e., not shift invariant. To address this shortcoming, we introduce novel data-adaptive designs for each of the modules in ViTs, such as tokenization, self-attention, patch merging, and positional encoding. With our proposed modules, we achieve true shift-equivariance on four well-established ViTs, namely, Swin, SwinV2, CvT, and MViTv2. Empirically, we evaluate the proposed adaptive models on image classification and semantic segmentation tasks. These models achieve competitive performance across three different datasets while maintaining 100% shift consistency.

Vaishnavi Subramanian, Tanveer Syeda-Mahmood, Minh N. Do

The availability of multi-modality datasets provides a unique opportunity to characterize the same object of interest using multiple viewpoints more comprehensively. In this work, we investigate the use of canonical correlation analysis (CCA) and penalized variants of CCA (pCCA) for the fusion of two modalities. We study a simple graphical model for the generation of two-modality data. We analytically show that, with known model parameters, posterior mean estimators that jointly use both modalities outperform arbitrary linear mixing of single modality posterior estimators in latent variable prediction. Penalized extensions of CCA (pCCA) that incorporate domain knowledge can discover correlations with high-dimensional, low-sample data, whereas traditional CCA is inapplicable. To facilitate the generation of multi-dimensional embeddings with pCCA, we propose two matrix deflation schemes that enforce desirable properties exhibited by CCA. We propose a two-stage prediction pipeline using pCCA embeddings generated with deflation for latent variable prediction by combining all the above. On simulated data, our proposed model drastically reduces the mean-squared error in latent variable prediction. When applied to publicly available histopathology data and RNA-sequencing data from The Cancer Genome Atlas (TCGA) breast cancer patients, our model can outperform principal components analysis (PCA) embeddings of the same dimension in survival prediction.

Qian Jiang, Xiaofan Zhang, Deming Chen et al.

In hardware-aware Differentiable Neural Architecture Search (DNAS), it is challenging to compute gradients of hardware metrics to perform architecture search. Existing works rely on linear approximations with limited support to customized hardware accelerators. In this work, we propose End-to-end Hardware-aware DNAS (EH-DNAS), a seamless integration of end-to-end hardware benchmarking, and fully automated DNAS to deliver hardware-efficient deep neural networks on various platforms, including Edge GPUs, Edge TPUs, Mobile CPUs, and customized accelerators. Given a desired hardware platform, we propose to learn a differentiable model predicting the end-to-end hardware performance of neural network architectures for DNAS. We also introduce E2E-Perf, an end-to-end hardware benchmarking tool for customized accelerators. Experiments on CIFAR10 and ImageNet show that EH-DNAS improves the hardware performance by an average of $1.4\times$ on customized accelerators and $1.6\times$ on existing hardware processors while maintaining the classification accuracy.

Renan A. Rojas-Gomez, Raymond A. Yeh, Minh N. Do et al.

Despite unconditional feature inversion being the foundation of many image synthesis applications, training an inverter demands a high computational budget, large decoding capacity and imposing conditions such as autoregressive priors. To address these limitations, we propose the use of adversarially robust representations as a perceptual primitive for feature inversion. We train an adversarially robust encoder to extract disentangled and perceptually-aligned image representations, making them easily invertible. By training a simple generator with the mirror architecture of the encoder, we achieve superior reconstruction quality and generalization over standard models. Based on this, we propose an adversarially robust autoencoder and demonstrate its improved performance on style transfer, image denoising and anomaly detection tasks. Compared to recent ImageNet feature inversion methods, our model attains improved performance with significantly less complexity.

Vaishnavi Subramanian, Tanveer Syeda-Mahmood, Minh N. Do

Effective understanding of a disease such as cancer requires fusing multiple sources of information captured across physical scales by multimodal data. In this work, we propose a novel feature embedding module that derives from canonical correlation analyses to account for intra-modality and inter-modality correlations. Experiments on simulated and real data demonstrate how our proposed module can learn well-correlated multi-dimensional embeddings. These embeddings perform competitively on one-year survival classification of TCGA-BRCA breast cancer patients, yielding average F1 scores up to 58.69% under 5-fold cross-validation.

Vaishnavi Subramanian, Minh N. Do, Tanveer Syeda-Mahmood

Lung cancer has a high rate of recurrence in early-stage patients. Predicting the post-surgical recurrence in lung cancer patients has traditionally been approached using single modality information of genomics or radiology images. We investigate the potential of multimodal fusion for this task. By combining computed tomography (CT) images and genomics, we demonstrate improved prediction of recurrence using linear Cox proportional hazards models with elastic net regularization. We work on a recent non-small cell lung cancer (NSCLC) radiogenomics dataset of 130 patients and observe an increase in concordance-index values of up to 10%. Employing non-linear methods from the neural network literature, such as multi-layer perceptrons and visual-question answering fusion modules, did not improve performance consistently. This indicates the need for larger multimodal datasets and fusion techniques better adapted to this biological setting.

Ramanpreet Singh Pahwa, Kennard Yanting Chan, Jiamin Bai et al.

Advances in Unmanned Aerial Vehicle (UAV) opens venues for application such as tunnel inspection. Owing to its versatility to fly inside the tunnels, it can quickly identify defects and potential problems related to safety. However, long tunnels, especially with repetitive or uniform structures pose a significant problem for UAV navigation. Furthermore, post-processing visual data from the camera mounted on the UAV is required to generate useful information for the inspection task. In this work, we design a UAV with a single rotating camera to accomplish the task. Compared to other platforms, our solution can fit the stringent requirement for tunnel inspection, in terms of battery life, size and weight. While the current state-of-the-art can estimate camera pose and 3D geometry from a sequence of images, they assume large overlap, small rotational motion, and many distinct matching points between images. These assumptions severely limit their effectiveness in tunnel-like scenarios where the camera has erratic or large rotational motion, such as the one mounted on the UAV. This paper presents a novel solution which exploits Structure-from-Motion, Bundle Adjustment, and available geometry priors to robustly estimate camera pose and automatically reconstruct a fully-dense 3D scene using the least possible number of images in various challenging tunnel-like environments. We validate our system with both Virtual Reality application and experimentation with a real dataset. The results demonstrate that the proposed reconstruction along with texture mapping allows for remote navigation and inspection of tunnel-like environments, even those which are inaccessible for humans.

Khoi-Nguyen C. Mac, Dhiraj Joshi, Raymond A. Yeh et al.

Fine-grained action detection is an important task with numerous applications in robotics and human-computer interaction. Existing methods typically utilize a two-stage approach including extraction of local spatio-temporal features followed by temporal modeling to capture long-term dependencies. While most recent papers have focused on the latter (long-temporal modeling), here, we focus on producing features capable of modeling fine-grained motion more efficiently. We propose a novel locally-consistent deformable convolution, which utilizes the change in receptive fields and enforces a local coherency constraint to capture motion information effectively. Our model jointly learns spatio-temporal features (instead of using independent spatial and temporal streams). The temporal component is learned from the feature space instead of pixel space, e.g. optical flow. The produced features can be flexibly used in conjunction with other long-temporal modeling networks, e.g. ST-CNN, DilatedTCN, and ED-TCN. Overall, our proposed approach robustly outperforms the original long-temporal models on two fine-grained action datasets: 50 Salads and GTEA, achieving F1 scores of 80.22% and 75.39% respectively.

Ramanpreet Singh Pahwa, Wei Kiat Leong, Shaohui Foong et al.

Traditional image stitching algorithms use transforms such as homography to combine different views of a scene. They usually work well when the scene is planar or when the camera is only rotated, keeping its position static. This severely limits their use in real world scenarios where an unmanned aerial vehicle (UAV) potentially hovers around and flies in an enclosed area while rotating to capture a video sequence. We utilize known scene geometry along with recorded camera trajectory to create cylindrical images captured in a given environment such as a tunnel where the camera rotates around its center. The captured images of the inner surface of the given scene are combined to create a composite panoramic image that is textured onto a 3D geometrical object in Unity graphical engine to create an immersive environment for end users.

Benjamin Chidester, Minh N. Do, Jian Ma

Performance of neural networks can be significantly improved by encoding known invariance for particular tasks. Many image classification tasks, such as those related to cellular imaging, exhibit invariance to rotation. We present a novel scheme using the magnitude response of the 2D-discrete-Fourier transform (2D-DFT) to encode rotational invariance in neural networks, along with a new, efficient convolutional scheme for encoding rotational equivariance throughout convolutional layers. We implemented this scheme for several image classification tasks and demonstrated improved performance, in terms of classification accuracy, time required to train the model, and robustness to hyperparameter selection, over a standard CNN and another state-of-the-art method.

Raymond A. Yeh, Jinjun Xiong, Wen-mei W. Hwu et al.

Textual grounding is an important but challenging task for human-computer interaction, robotics and knowledge mining. Existing algorithms generally formulate the task as selection from a set of bounding box proposals obtained from deep net based systems. In this work, we demonstrate that we can cast the problem of textual grounding into a unified framework that permits efficient search over all possible bounding boxes. Hence, the method is able to consider significantly more proposals and doesn't rely on a successful first stage hypothesizing bounding box proposals. Beyond, we demonstrate that the trained parameters of our model can be used as word-embeddings which capture spatial-image relationships and provide interpretability. Lastly, at the time of submission, our approach outperformed the current state-of-the-art methods on the Flickr 30k Entities and the ReferItGame dataset by 3.08% and 7.77% respectively.

Raymond A. Yeh, Minh N. Do, Alexander G. Schwing

Textual grounding, i.e., linking words to objects in images, is a challenging but important task for robotics and human-computer interaction. Existing techniques benefit from recent progress in deep learning and generally formulate the task as a supervised learning problem, selecting a bounding box from a set of possible options. To train these deep net based approaches, access to a large-scale datasets is required, however, constructing such a dataset is time-consuming and expensive. Therefore, we develop a completely unsupervised mechanism for textual grounding using hypothesis testing as a mechanism to link words to detected image concepts. We demonstrate our approach on the ReferIt Game dataset and the Flickr30k data, outperforming baselines by 7.98% and 6.96% respectively.

Ramanpreet Singh Pahwa, Tian Tsong Ng, Minh N. Do

3D object proposals, quickly detected regions in a 3D scene that likely contain an object of interest, are an effective approach to improve the computational efficiency and accuracy of the object detection framework. In this work, we propose a novel online method that uses our previously developed 3D object proposals, in a RGB-D video sequence, to match and track static objects in the scene using shape matching. Our main observation is that depth images provide important information about the geometry of the scene that is often ignored in object matching techniques. Our method takes less than a second in MATLAB on the UW-RGBD scene dataset on a single thread CPU and thus, has potential to be used in low-power chips in Unmanned Aerial Vehicles (UAVs), quadcopters, and drones.

Ramanpreet Singh Pahwa, Jiangbo Lu, Nianjuan Jiang et al.

2D object proposals, quickly detected regions in an image that likely contain an object of interest, are an effective approach for improving the computational efficiency and accuracy of object detection in color images. In this work, we propose a novel online method that generates 3D object proposals in a RGB-D video sequence. Our main observation is that depth images provide important information about the geometry of the scene. Diverging from the traditional goal of 2D object proposals to provide a high recall (lots of 2D bounding boxes near potential objects), we aim for precise 3D proposals. We leverage on depth information per frame and multi-view scene information to obtain accurate 3D object proposals. Using efficient but robust registration enables us to combine multiple frames of a scene in near real time and generate 3D bounding boxes for potential 3D regions of interest. Using standard metrics, such as Precision-Recall curves and F-measure, we show that the proposed approach is significantly more accurate than the current state-of-the-art techniques. Our online approach can be integrated into SLAM based video processing for quick 3D object localization. Our method takes less than a second in MATLAB on the UW-RGBD scene dataset on a single thread CPU and thus, has potential to be used in low-power chips in Unmanned Aerial Vehicles (UAVs), quadcopters, and drones.

Ramanpreet Singh Pahwa, Minh N. Do, Tian Tsong Ng et al.

Depth sensing devices have created various new applications in scientific and commercial research with the advent of Microsoft Kinect and PMD (Photon Mixing Device) cameras. Most of these applications require the depth cameras to be pre-calibrated. However, traditional calibration methods using a checkerboard do not work very well for depth cameras due to the low image resolution. In this paper, we propose a depth calibration scheme which excels in estimating camera calibration parameters when only a handful of corners and calibration images are available. We exploit the noise properties of PMD devices to denoise depth measurements and perform camera calibration using the denoised depth as an additional set of measurements. Our synthetic and real experiments show that our depth denoising and depth based calibration scheme provides significantly better results than traditional calibration methods.

Johann A. Bengua, Ho N. Phien, Hoang D. Tuan et al.

This paper introduces matrix product state (MPS) decomposition as a new and systematic method to compress multidimensional data represented by higher-order tensors. It solves two major bottlenecks in tensor compression: computation and compression quality. Regardless of tensor order, MPS compresses tensors to matrices of moderate dimension which can be used for classification. Mainly based on a successive sequence of singular value decompositions (SVD), MPS is quite simple to implement and arrives at the global optimal matrix, bypassing local alternating optimization, which is not only computationally expensive but cannot yield the global solution. Benchmark results show that MPS can achieve better classification performance with favorable computation cost compared to other tensor compression methods.

Raymond A. Yeh, Chen Chen, Teck Yian Lim et al.

Semantic image inpainting is a challenging task where large missing regions have to be filled based on the available visual data. Existing methods which extract information from only a single image generally produce unsatisfactory results due to the lack of high level context. In this paper, we propose a novel method for semantic image inpainting, which generates the missing content by conditioning on the available data. Given a trained generative model, we search for the closest encoding of the corrupted image in the latent image manifold using our context and prior losses. This encoding is then passed through the generative model to infer the missing content. In our method, inference is possible irrespective of how the missing content is structured, while the state-of-the-art learning based method requires specific information about the holes in the training phase. Experiments on three datasets show that our method successfully predicts information in large missing regions and achieves pixel-level photorealism, significantly outperforming the state-of-the-art methods.

Johann A. Bengua, Hoang D. Tuan, Ho N. Phien et al.

This paper proposes a novel framework called concatenated image completion via tensor augmentation and completion (ICTAC), which recovers missing entries of color images with high accuracy. Typical images are second- or third-order tensors (2D/3D) depending if they are grayscale or color, hence tensor completion algorithms are ideal for their recovery. The proposed framework performs image completion by concatenating copies of a single image that has missing entries into a third-order tensor, applying a dimensionality augmentation technique to the tensor, utilizing a tensor completion algorithm for recovering its missing entries, and finally extracting the recovered image from the tensor. The solution relies on two key components that have been recently proposed to take advantage of the tensor train (TT) rank: A tensor augmentation tool called ket augmentation (KA) that represents a low-order tensor by a higher-order tensor, and the algorithm tensor completion by parallel matrix factorization via tensor train (TMac-TT), which has been demonstrated to outperform state-of-the-art tensor completion algorithms. Simulation results for color image recovery show the clear advantage of our framework against current state-of-the-art tensor completion algorithms.

Seungryong Kim, Dongbo Min, Bumsub Ham et al.

Establishing dense correspondences between multiple images is a fundamental task in many applications. However, finding a reliable correspondence in multi-modal or multi-spectral images still remains unsolved due to their challenging photometric and geometric variations. In this paper, we propose a novel dense descriptor, called dense adaptive self-correlation (DASC), to estimate multi-modal and multi-spectral dense correspondences. Based on an observation that self-similarity existing within images is robust to imaging modality variations, we define the descriptor with a series of an adaptive self-correlation similarity measure between patches sampled by a randomized receptive field pooling, in which a sampling pattern is obtained using a discriminative learning. The computational redundancy of dense descriptors is dramatically reduced by applying fast edge-aware filtering. Furthermore, in order to address geometric variations including scale and rotation, we propose a geometry-invariant DASC (GI-DASC) descriptor that effectively leverages the DASC through a superpixel-based representation. For a quantitative evaluation of the GI-DASC, we build a novel multi-modal benchmark as varying photometric and geometric conditions. Experimental results demonstrate the outstanding performance of the DASC and GI-DASC in many cases of multi-modal and multi-spectral dense correspondences.

Yu Zhang, Xiu-shen Wei, Jianxin Wu et al.

In this paper, we categorize fine-grained images without using any object / part annotation neither in the training nor in the testing stage, a step towards making it suitable for deployments. Fine-grained image categorization aims to classify objects with subtle distinctions. Most existing works heavily rely on object / part detectors to build the correspondence between object parts by using object or object part annotations inside training images. The need for expensive object annotations prevents the wide usage of these methods. Instead, we propose to select useful parts from multi-scale part proposals in objects, and use them to compute a global image representation for categorization. This is specially designed for the annotation-free fine-grained categorization task, because useful parts have shown to play an important role in existing annotation-dependent works but accurate part detectors can be hardly acquired. With the proposed image representation, we can further detect and visualize the key (most discriminative) parts in objects of different classes. In the experiment, the proposed annotation-free method achieves better accuracy than that of state-of-the-art annotation-free and most existing annotation-dependent methods on two challenging datasets, which shows that it is not always necessary to use accurate object / part annotations in fine-grained image categorization.

Johann A. Bengua, Ho N. Phien, Hoang D. Tuan et al.

This paper introduces matrix product state (MPS) decomposition as a computational tool for extracting features of multidimensional data represented by higher-order tensors. Regardless of tensor order, MPS extracts its relevant features to the so-called core tensor of maximum order three which can be used for classification. Mainly based on a successive sequence of singular value decompositions (SVD), MPS is quite simple to implement without any recursive procedure needed for optimizing local tensors. Thus, it leads to substantial computational savings compared to other tensor feature extraction methods such as higher-order orthogonal iteration (HOOI) underlying the Tucker decomposition (TD). Benchmark results show that MPS can reduce significantly the feature space of data while achieving better classification performance compared to HOOI.

Viet-Anh Nguyen, Jiangbo Lu, Shengkui Zhao et al.

This paper presents an Immersive Telepresence system for Entertainment and Meetings (ITEM). The system aims to provide a radically new video communication experience by seamlessly merging participants into the same virtual space to allow a natural interaction among them and shared collaborative contents. With the goal to make a scalable, flexible system for various business solutions as well as easily accessible by massive consumers, we address the challenges in the whole pipeline of media processing, communication, and displaying in our design and realization of such a system. Particularly, in this paper we focus on the system aspects that maximize the end-user experience, optimize the system and network resources, and enable various teleimmersive application scenarios. In addition, we also present a few key technologies, i.e. fast object-based video coding for real world data and spatialized audio capture and 3D sound localization for group teleconferencing. Our effort is to investigate and optimize the key system components and provide an efficient end-to-end optimization and integration by considering user needs and preferences. Extensive experiments show the developed system runs reliably and comfortably in real time with a minimal setup requirement (e.g. a webcam and/or a depth camera, an optional microphone array, a laptop/desktop connected to the public Internet) for teleimmersive communication. With such a really minimal deployment requirement, we present a variety of interesting applications and user experiences created by ITEM.