Binh-Son Hua

Tuan Duc Ngo, Binh-Son Hua, Khoi Nguyen

Existing 3D instance segmentation methods are predominated by the bottom-up design -- manually fine-tuned algorithm to group points into clusters followed by a refinement network. However, by relying on the quality of the clusters, these methods generate susceptible results when (1) nearby objects with the same semantic class are packed together, or (2) large objects with loosely connected regions. To address these limitations, we introduce ISBNet, a novel cluster-free method that represents instances as kernels and decodes instance masks via dynamic convolution. To efficiently generate high-recall and discriminative kernels, we propose a simple strategy named Instance-aware Farthest Point Sampling to sample candidates and leverage the local aggregation layer inspired by PointNet++ to encode candidate features. Moreover, we show that predicting and leveraging the 3D axis-aligned bounding boxes in the dynamic convolution further boosts performance. Our method set new state-of-the-art results on ScanNetV2 (55.9), S3DIS (60.8), and STPLS3D (49.2) in terms of AP and retains fast inference time (237ms per scene on ScanNetV2). The source code and trained models are available at https://github.com/VinAIResearch/ISBNet.

Khoi D. Nguyen, Quoc-Huy Tran, Khoi Nguyen et al.

We present a novel method for few-shot video classification, which performs appearance and temporal alignments. In particular, given a pair of query and support videos, we conduct appearance alignment via frame-level feature matching to achieve the appearance similarity score between the videos, while utilizing temporal order-preserving priors for obtaining the temporal similarity score between the videos. Moreover, we introduce a few-shot video classification framework that leverages the above appearance and temporal similarity scores across multiple steps, namely prototype-based training and testing as well as inductive and transductive prototype refinement. To the best of our knowledge, our work is the first to explore transductive few-shot video classification. Extensive experiments on both Kinetics and Something-Something V2 datasets show that both appearance and temporal alignments are crucial for datasets with temporal order sensitivity such as Something-Something V2. Our approach achieves similar or better results than previous methods on both datasets. Our code is available at https://github.com/VinAIResearch/fsvc-ata.

Phuoc-Hieu Le, Quynh Le, Rang Nguyen et al.

High dynamic range (HDR) imaging is an indispensable technique in modern photography. Traditional methods focus on HDR reconstruction from multiple images, solving the core problems of image alignment, fusion, and tone mapping, yet having a perfect solution due to ghosting and other visual artifacts in the reconstruction. Recent attempts at single-image HDR reconstruction show a promising alternative: by learning to map pixel values to their irradiance using a neural network, one can bypass the align-and-merge pipeline completely yet still obtain a high-quality HDR image. In this work, we propose a weakly supervised learning method that inverts the physical image formation process for HDR reconstruction via learning to generate multiple exposures from a single image. Our neural network can invert the camera response to reconstruct pixel irradiance before synthesizing multiple exposures and hallucinating details in under- and over-exposed regions from a single input image. To train the network, we propose a representation loss, a reconstruction loss, and a perceptual loss applied on pairs of under- and over-exposure images and thus do not require HDR images for training. Our experiments show that our proposed model can effectively reconstruct HDR images. Our qualitative and quantitative results show that our method achieves state-of-the-art performance on the DrTMO dataset. Our code is available at https://github.com/VinAIResearch/single_image_hdr.

Ngoc-Vuong Ho, Tan Nguyen, Gia-Han Diep et al.

Medical image analysis using deep learning has recently been prevalent, showing great performance for various downstream tasks including medical image segmentation and its sibling, volumetric image segmentation. Particularly, a typical volumetric segmentation network strongly relies on a voxel grid representation which treats volumetric data as a stack of individual voxel `slices', which allows learning to segment a voxel grid to be as straightforward as extending existing image-based segmentation networks to the 3D domain. However, using a voxel grid representation requires a large memory footprint, expensive test-time and limiting the scalability of the solutions. In this paper, we propose Point-Unet, a novel method that incorporates the efficiency of deep learning with 3D point clouds into volumetric segmentation. Our key idea is to first predict the regions of interest in the volume by learning an attentional probability map, which is then used for sampling the volume into a sparse point cloud that is subsequently segmented using a point-based neural network. We have conducted the experiments on the medical volumetric segmentation task with both a small-scale dataset Pancreas and large-scale datasets BraTS18, BraTS19, and BraTS20 challenges. A comprehensive benchmark on different metrics has shown that our context-aware Point-Unet robustly outperforms the SOTA voxel-based networks at both accuracies, memory usage during training, and time consumption during testing. Our code is available at https://github.com/VinAIResearch/Point-Unet.

Jaeyeon Kim, Binh-Son Hua, Duc Thanh Nguyen et al.

In this paper, we propose a new method for mapping a 3D point cloud to the latent space of a 3D generative adversarial network. Our generative model for 3D point clouds is based on SP-GAN, a state-of-the-art sphere-guided 3D point cloud generator. We derive an efficient way to encode an input 3D point cloud to the latent space of the SP-GAN. Our point cloud encoder can resolve the point ordering issue during inversion, and thus can determine the correspondences between points in the generated 3D point cloud and those in the canonical sphere used by the generator. We show that our method outperforms previous GAN inversion methods for 3D point clouds, achieving state-of-the-art results both quantitatively and qualitatively. Our code is available at https://github.com/hkust-vgd/point_inverter.

Bach Tran, Binh-Son Hua, Anh Tuan Tran et al.

Recently, great progress has been made in 3D deep learning with the emergence of deep neural networks specifically designed for 3D point clouds. These networks are often trained from scratch or from pre-trained models learned purely from point cloud data. Inspired by the success of deep learning in the image domain, we devise a novel pre-training technique for better model initialization by utilizing the multi-view rendering of the 3D data. Our pre-training is self-supervised by a local pixel/point level correspondence loss computed from perspective projection and a global image/point cloud level loss based on knowledge distillation, thus effectively improving upon popular point cloud networks, including PointNet, DGCNN and SR-UNet. These improved models outperform existing state-of-the-art methods on various datasets and downstream tasks. We also analyze the benefits of synthetic and real data for pre-training, and observe that pre-training on synthetic data is also useful for high-level downstream tasks. Code and pre-trained models are available at https://github.com/VinAIResearch/selfsup_pcd.

Yingshu Chen, Tuan-Anh Vu, Ka-Chun Shum et al.

Architectural photography is a genre of photography that focuses on capturing a building or structure in the foreground with dramatic lighting in the background. Inspired by recent successes in image-to-image translation methods, we aim to perform style transfer for architectural photographs. However, the special composition in architectural photography poses great challenges for style transfer in this type of photographs. Existing neural style transfer methods treat the architectural images as a single entity, which would generate mismatched chrominance and destroy geometric features of the original architecture, yielding unrealistic lighting, wrong color rendition, and visual artifacts such as ghosting, appearance distortion, or color mismatching. In this paper, we specialize a neural style transfer method for architectural photography. Our method addresses the composition of the foreground and background in an architectural photograph in a two-branch neural network that separately considers the style transfer of the foreground and the background, respectively. Our method comprises a segmentation module, a learning-based image-to-image translation module, and an image blending optimization module. We trained our image-to-image translation neural network with a new dataset of unconstrained outdoor architectural photographs captured at different magic times of a day, utilizing additional semantic information for better chrominance matching and geometry preservation. Our experiments show that our method can produce photorealistic lighting and color rendition on both the foreground and background, and outperforms general image-to-image translation and arbitrary style transfer baselines quantitatively and qualitatively. Our code and data are available at https://github.com/hkust-vgd/architectural_style_transfer.

Tuan Duc Ngo, Binh-Son Hua, Khoi Nguyen

Instance segmentation on 3D point clouds (3DIS) is a longstanding challenge in computer vision, where state-of-the-art methods are mainly based on full supervision. As annotating ground truth dense instance masks is tedious and expensive, solving 3DIS with weak supervision has become more practical. In this paper, we propose GaPro, a new instance segmentation for 3D point clouds using axis-aligned 3D bounding box supervision. Our two-step approach involves generating pseudo labels from box annotations and training a 3DIS network with the resulting labels. Additionally, we employ the self-training strategy to improve the performance of our method further. We devise an effective Gaussian Process to generate pseudo instance masks from the bounding boxes and resolve ambiguities when they overlap, resulting in pseudo instance masks with their uncertainty values. Our experiments show that GaPro outperforms previous weakly supervised 3D instance segmentation methods and has competitive performance compared to state-of-the-art fully supervised ones. Furthermore, we demonstrate the robustness of our approach, where we can adapt various state-of-the-art fully supervised methods to the weak supervision task by using our pseudo labels for training. The source code and trained models are available at https://github.com/VinAIResearch/GaPro.

Tan Nguyen, Binh-Son Hua, Ngan Le

Medical image segmentation has been so far achieving promising results with Convolutional Neural Networks (CNNs). However, it is arguable that in traditional CNNs, its pooling layer tends to discard important information such as positions. Moreover, CNNs are sensitive to rotation and affine transformation. Capsule network is a data-efficient network design proposed to overcome such limitations by replacing pooling layers with dynamic routing and convolutional strides, which aims to preserve the part-whole relationships. Capsule network has shown a great performance in image recognition and natural language processing, but applications for medical image segmentation, particularly volumetric image segmentation, has been limited. In this work, we propose 3D-UCaps, a 3D voxel-based Capsule network for medical volumetric image segmentation. We build the concept of capsules into a CNN by designing a network with two pathways: the first pathway is encoded by 3D Capsule blocks, whereas the second pathway is decoded by 3D CNNs blocks. 3D-UCaps, therefore inherits the merits from both Capsule network to preserve the spatial relationship and CNNs to learn visual representation. We conducted experiments on various datasets to demonstrate the robustness of 3D-UCaps including iSeg-2017, LUNA16, Hippocampus, and Cardiac, where our method outperforms previous Capsule networks and 3D-Unets.

Duong H. Le, Khoi D. Nguyen, Khoi Nguyen et al.

In this work, we propose to use out-of-distribution samples, i.e., unlabeled samples coming from outside the target classes, to improve few-shot learning. Specifically, we exploit the easily available out-of-distribution samples to drive the classifier to avoid irrelevant features by maximizing the distance from prototypes to out-of-distribution samples while minimizing that of in-distribution samples (i.e., support, query data). Our approach is simple to implement, agnostic to feature extractors, lightweight without any additional cost for pre-training, and applicable to both inductive and transductive settings. Extensive experiments on various standard benchmarks demonstrate that the proposed method consistently improves the performance of pretrained networks with different architectures.

Ka Chun Shum, Hong-Wing Pang, Binh-Son Hua et al.

In this paper, we address the problem of conditional scene decoration for 360-degree images. Our method takes a 360-degree background photograph of an indoor scene and generates decorated images of the same scene in the panorama view. To do this, we develop a 360-aware object layout generator that learns latent object vectors in the 360-degree view to enable a variety of furniture arrangements for an input 360-degree background image. We use this object layout to condition a generative adversarial network to synthesize images of an input scene. To further reinforce the generation capability of our model, we develop a simple yet effective scene emptier that removes the generated furniture and produces an emptied scene for our model to learn a cyclic constraint. We train the model on the Structure3D dataset and show that our model can generate diverse decorations with controllable object layout. Our method achieves state-of-the-art performance on the Structure3D dataset and generalizes well to the Zillow indoor scene dataset. Our user study confirms the immersive experiences provided by the realistic image quality and furniture layout in our generation results. Our implementation will be made available.

Hong-Wing Pang, Binh-Son Hua, Sai-Kit Yeung

In recent years, there has been increasing interest in applying stylization on 3D scenes from a reference style image, in particular onto neural radiance fields (NeRF). While performing stylization directly on NeRF guarantees appearance consistency over arbitrary novel views, it is a challenging problem to guide the transfer of patterns from the style image onto different parts of the NeRF scene. In this work, we propose a stylization framework for NeRF based on local style transfer. In particular, we use a hash-grid encoding to learn the embedding of the appearance and geometry components, and show that the mapping defined by the hash table allows us to control the stylization to a certain extent. Stylization is then achieved by optimizing the appearance branch while keeping the geometry branch fixed. To support local style transfer, we propose a new loss function that utilizes a segmentation network and bipartite matching to establish region correspondences between the style image and the content images obtained from volume rendering. Our experiments show that our method yields plausible stylization results with novel view synthesis while having flexible controllability via manipulating and customizing the region correspondences.

Yingshu Chen, Guocheng Shao, Ka Chun Shum et al.

Modern artificial intelligence offers a novel and transformative approach to creating digital art across diverse styles and modalities like images, videos and 3D data, unleashing the power of creativity and revolutionizing the way that we perceive and interact with visual content. This paper reports on recent advances in stylized 3D asset creation and manipulation with the expressive power of neural networks. We establish a taxonomy for neural stylization, considering crucial design choices such as scene representation, guidance data, optimization strategies, and output styles. Building on such taxonomy, our survey first revisits the background of neural stylization on 2D images, and then presents in-depth discussions on recent neural stylization methods for 3D data, accompanied by a benchmark evaluating selected mesh and neural field stylization methods. Based on the insights gained from the survey, we highlight the practical significance, open challenges, future research, and potential impacts of neural stylization, which facilitates researchers and practitioners to navigate the rapidly evolving landscape of 3D content creation using modern artificial intelligence.

Ka Chun Shum, Jaeyeon Kim, Binh-Son Hua et al.

Neural radiance field is an emerging rendering method that generates high-quality multi-view consistent images from a neural scene representation and volume rendering. Although neural radiance field-based techniques are robust for scene reconstruction, their ability to add or remove objects remains limited. This paper proposes a new language-driven approach for object manipulation with neural radiance fields through dataset updates. Specifically, to insert a new foreground object represented by a set of multi-view images into a background radiance field, we use a text-to-image diffusion model to learn and generate combined images that fuse the object of interest into the given background across views. These combined images are then used for refining the background radiance field so that we can render view-consistent images containing both the object and the background. To ensure view consistency, we propose a dataset updates strategy that prioritizes radiance field training with camera views close to the already-trained views prior to propagating the training to remaining views. We show that under the same dataset updates strategy, we can easily adapt our method for object insertion using data from text-to-3D models as well as object removal. Experimental results show that our method generates photorealistic images of the edited scenes, and outperforms state-of-the-art methods in 3D reconstruction and neural radiance field blending.

Tuan-Anh Vu, Srinjay Sarkar, Zhiyuan Zhang et al.

Data augmentation is a powerful technique to enhance the performance of a deep learning task but has received less attention in 3D deep learning. It is well known that when 3D shapes are sparsely represented with low point density, the performance of the downstream tasks drops significantly. This work explores test-time augmentation (TTA) for 3D point clouds. We are inspired by the recent revolution of learning implicit representation and point cloud upsampling, which can produce high-quality 3D surface reconstruction and proximity-to-surface, respectively. Our idea is to leverage the implicit field reconstruction or point cloud upsampling techniques as a systematic way to augment point cloud data. Mainly, we test both strategies by sampling points from the reconstructed results and using the sampled point cloud as test-time augmented data. We show that both strategies are effective in improving accuracy. We observed that point cloud upsampling for test-time augmentation can lead to more significant performance improvement on downstream tasks such as object classification and segmentation on the ModelNet40, ShapeNet, ScanObjectNN, and SemanticKITTI datasets, especially for sparse point clouds.

Hubert Kompanowski, Varun Jampani, Aaryaman Vasishta et al.

Multiview diffusion models have rapidly emerged as a powerful tool for content creation with spatial consistency across viewpoints, offering rich visual realism without requiring explicit geometry and appearance representation. However, compared to meshes or radiance fields, existing multiview diffusion models offer limited appearance manipulation, particularly in terms of material, texture, or style. In this paper, we present a lightweight adaptation technique for appearance transfer in multiview diffusion models. Our method learns to combine object identity from an input image with appearance cues rendered in a separate reference image, producing multi-view-consistent output that reflects the desired materials, textures, or styles. This allows explicit specification of appearance parameters at generation time while preserving the underlying object geometry and view coherence. We leverage three diffusion denoising processes responsible for generating the original object, the reference, and the target images, and perform reverse sampling to aggregate a small subset of layer-wise self-attention features from the object and the reference to influence the target generation. Our method requires only a few training examples to introduce appearance awareness to pretrained multiview models. The experiments show that our method provides a simple yet effective way toward multiview generation with diverse appearance, advocating the adoption of implicit generative 3D representations in practice.

Phuc Pham, Uy Dieu Tran, Binh-Son Hua et al.

Realistic and efficient 3D garment generation remains a longstanding challenge in computer vision and digital fashion. Existing methods typically rely on large vision- language models to produce serialized representations of 2D sewing patterns, which are then transformed into simulation-ready 3D meshes using garment modeling framework such as GarmentCode. Although these approaches yield high-quality results, they often suffer from slow inference times, ranging from 30 seconds to a minute. In this work, we introduce SwiftTailor, a novel two-stage framework that unifies sewing-pattern reasoning and geometry-based mesh synthesis through a compact geometry image representation. SwiftTailor comprises two lightweight modules: PatternMaker, an efficient vision-language model that predicts sewing patterns from diverse input modalities, and GarmentSewer, an efficient dense prediction transformer that converts these patterns into a novel Garment Geometry Image, encoding the 3D surface of all garment panels in a unified UV space. The final 3D mesh is reconstructed through an efficient inverse mapping process that incorporates remeshing and dynamic stitching algorithms to directly assemble the garment, thereby amortizing the cost of physical simulation. Extensive experiments on the Multimodal GarmentCodeData demonstrate that SwiftTailor achieves state-of-the-art accuracy and visual fidelity while significantly reducing inference time. This work offers a scalable, interpretable, and high-performance solution for next-generation 3D garment generation.

Tuan-Anh Vu, Hai Nguyen-Truong, Ziqiang Zheng et al.

Glass is a prevalent material among solid objects in everyday life, yet segmentation methods struggle to distinguish it from opaque materials due to its transparency and reflection. While it is known that human perception relies on boundary and reflective-object features to distinguish glass objects, the existing literature has not yet sufficiently captured both properties when handling transparent objects. Hence, we propose incorporating both of these powerful visual cues via the Boundary Feature Enhancement and Reflection Feature Enhancement modules in a mutually beneficial way. Our proposed framework, TransCues, is a pyramidal transformer encoder-decoder architecture to segment transparent objects. We empirically show that these two modules can be used together effectively, improving overall performance across various benchmark datasets, including glass object semantic segmentation, mirror object semantic segmentation, and generic segmentation datasets. Our method outperforms the state-of-the-art by a large margin, achieving +4.2% mIoU on Trans10K-v2, +5.6% mIoU on MSD, +10.1% mIoU on RGBD-Mirror, +13.1% mIoU on TROSD, and +8.3% mIoU on Stanford2D3D, showing the effectiveness of our method against glass objects.

Tuan-Anh Vu, Duc Thanh Nguyen, Qing Guo et al.

Text-to-image diffusion techniques have shown exceptional capability of producing high-quality images from text descriptions. This indicates that there exists a strong correlation between the visual and textual domains. In addition, text-image discriminative models such as CLIP excel in image labelling from text prompts, thanks to the rich and diverse information available from open concepts. In this paper, we leverage these technical advances to solve a challenging problem in computer vision: camouflaged instance segmentation. Specifically, we propose a method built upon a state-of-the-art diffusion model, empowered by open-vocabulary to learn multi-scale textual-visual features for camouflaged object representations. Such cross-domain representations are desirable in segmenting camouflaged objects where visual cues are subtle to distinguish the objects from the background, especially in segmenting novel objects which are not seen in training. We also develop technically supportive components to effectively fuse cross-domain features and engage relevant features towards respective foreground objects. We validate our method and compare it with existing ones on several benchmark datasets of camouflaged instance segmentation and generic open-vocabulary instance segmentation. Experimental results confirm the advances of our method over existing ones. We will publish our code and pre-trained models to support future research.

Tuan-Anh Vu, Duc Thanh Nguyen, Binh-Son Hua et al.

Object reconstruction from 3D point clouds has been a long-standing research problem in computer vision and computer graphics, and achieved impressive progress. However, reconstruction from time-varying point clouds (a.k.a. 4D point clouds) is generally overlooked. In this paper, we propose a new network architecture, namely RFNet-4D++, that jointly reconstructs objects and their motion flows from 4D point clouds. The key insight is simultaneously performing both tasks via learning of spatial and temporal features from a sequence of point clouds can leverage individual tasks, leading to improved overall performance. To prove this ability, we design a temporal vector field learning module using an unsupervised learning approach for flow estimation task, leveraged by supervised learning of spatial structures for object reconstruction. Extensive experiments and analyses on benchmark datasets validated the effectiveness and efficiency of our method. As shown in experimental results, our method achieves state-of-the-art performance on both flow estimation and object reconstruction while performing much faster than existing methods in both training and inference. Our code and data are available at https://github.com/hkust-vgd/RFNet-4D

Hong-Wing Pang, Yingshu Chen, Phuoc-Hieu Le et al.

Furnishing and rendering indoor scenes has been a long-standing task for interior design, where artists create a conceptual design for the space, build a 3D model of the space, decorate, and then perform rendering. Although the task is important, it is tedious and requires tremendous effort. In this paper, we introduce a new problem of domain-specific indoor scene image synthesis, namely neural scene decoration. Given a photograph of an empty indoor space and a list of decorations with layout determined by user, we aim to synthesize a new image of the same space with desired furnishing and decorations. Neural scene decoration can be applied to create conceptual interior designs in a simple yet effective manner. Our attempt to this research problem is a novel scene generation architecture that transforms an empty scene and an object layout into a realistic furnished scene photograph. We demonstrate the performance of our proposed method by comparing it with conditional image synthesis baselines built upon prevailing image translation approaches both qualitatively and quantitatively. We conduct extensive experiments to further validate the plausibility and aesthetics of our generated scenes. Our implementation is available at \url{https://github.com/hkust-vgd/neural_scene_decoration}.

Duc-Hai Pham, Tung Do, Phong Nguyen et al.

We propose SharpDepth, a novel approach to monocular metric depth estimation that combines the metric accuracy of discriminative depth estimation methods (e.g., Metric3D, UniDepth) with the fine-grained boundary sharpness typically achieved by generative methods (e.g., Marigold, Lotus). Traditional discriminative models trained on real-world data with sparse ground-truth depth can accurately predict metric depth but often produce over-smoothed or low-detail depth maps. Generative models, in contrast, are trained on synthetic data with dense ground truth, generating depth maps with sharp boundaries yet only providing relative depth with low accuracy. Our approach bridges these limitations by integrating metric accuracy with detailed boundary preservation, resulting in depth predictions that are both metrically precise and visually sharp. Our extensive zero-shot evaluations on standard depth estimation benchmarks confirm SharpDepth effectiveness, showing its ability to achieve both high depth accuracy and detailed representation, making it well-suited for applications requiring high-quality depth perception across diverse, real-world environments.

Hai Nguyen-Truong, E-Ro Nguyen, Tuan-Anh Vu et al.

Referring image segmentation is a challenging task that involves generating pixel-wise segmentation masks based on natural language descriptions. The complexity of this task increases with the intricacy of the sentences provided. Existing methods have relied mostly on visual features to generate the segmentation masks while treating text features as supporting components. However, this under-utilization of text understanding limits the model's capability to fully comprehend the given expressions. In this work, we propose a novel framework that specifically emphasizes object and context comprehension inspired by human cognitive processes through Vision-Aware Text Features. Firstly, we introduce a CLIP Prior module to localize the main object of interest and embed the object heatmap into the query initialization process. Secondly, we propose a combination of two components: Contextual Multimodal Decoder and Meaning Consistency Constraint, to further enhance the coherent and consistent interpretation of language cues with the contextual understanding obtained from the image. Our method achieves significant performance improvements on three benchmark datasets RefCOCO, RefCOCO+ and G-Ref. Project page: \url{https://vatex.hkustvgd.com/}.

Khoi Do, Duong Nguyen, Nam-Khanh Le et al.

In this study, we address the gradient-based domain generalization problem, where predictors aim for consistent gradient directions across different domains. Existing methods have two main challenges. First, minimization of gradient empirical distance or gradient inner products (GIP) leads to gradient fluctuations among domains, thereby hindering straightforward learning. Second, the direct application of gradient learning to the joint loss function can incur high computation overheads due to second-order derivative approximation. To tackle these challenges, we propose a new Pareto Optimality Gradient Matching (POGM) method. In contrast to existing methods that add gradient matching as regularization, we leverage gradient trajectories as collected data and apply independent training at the meta-learner. In the meta-update, we maximize GIP while limiting the learned gradient from deviating too far from the empirical risk minimization gradient trajectory. By doing so, the aggregate gradient can incorporate knowledge from all domains without suffering gradient fluctuation towards any particular domain. Experimental evaluations on datasets from DomainBed demonstrate competitive results yielded by POGM against other baselines while achieving computational efficiency.

Quang Trung Truong, Wong Yuk Kwan, Duc Thanh Nguyen et al.

Underwater video analysis, hampered by the dynamic marine environment and camera motion, remains a challenging task in computer vision. Existing training-free video generation techniques, learning motion dynamics on the frame-by-frame basis, often produce poor results with noticeable motion interruptions and misaligments. To address these issues, we propose AUTV, a framework for synthesizing marine video data with pixel-wise annotations. We demonstrate the effectiveness of this framework by constructing two video datasets, namely UTV, a real-world dataset comprising 2,000 video-text pairs, and SUTV, a synthetic video dataset including 10,000 videos with segmentation masks for marine objects. UTV provides diverse underwater videos with comprehensive annotations including appearance, texture, camera intrinsics, lighting, and animal behavior. SUTV can be used to improve underwater downstream tasks, which are demonstrated in video inpainting and video object segmentation.

Ka Chun Shum, Binh-Son Hua, Duc Thanh Nguyen et al.

Diffusion models have shown great promise in synthesizing visually appealing images. However, it remains challenging to condition the synthesis at a fine-grained level, for instance, synthesizing image pixels following some generic color pattern. Existing image synthesis methods often produce contents that fall outside the desired pixel conditions. To address this, we introduce a novel color alignment algorithm that confines the generative process in diffusion models within a given color pattern. Specifically, we project diffusion terms, either imagery samples or latent representations, into a conditional color space to align with the input color distribution. This strategy simplifies the prediction in diffusion models within a color manifold while still allowing plausible structures in generated contents, thus enabling the generation of diverse contents that comply with the target color pattern. Experimental results demonstrate our state-of-the-art performance in conditioning and controlling of color pixels, while maintaining on-par generation quality and diversity in comparison with regular diffusion models.

Khoi Do, Binh-Son Hua

Score distillation sampling is an effective technique to generate 3D models from text prompts, utilizing pre-trained large-scale text-to-image diffusion models as guidance. However, the produced 3D assets tend to be over-saturating, over-smoothing, with limited diversity. These issues are results from a reverse Kullback-Leibler (KL) divergence objective, which makes the optimization unstable and results in mode-seeking behavior. In this paper, we derive a bounded score distillation objective based on Jensen-Shannon divergence (JSD), which stabilizes the optimization process and produces high-quality 3D generation. JSD can match well generated and target distribution, therefore mitigating mode seeking. We provide a practical implementation of JSD by utilizing the theory of generative adversarial networks to define an approximate objective function for the generator, assuming the discriminator is well trained. By assuming the discriminator following a log-odds classifier, we propose a minority sampling algorithm to estimate the gradients of our proposed objective, providing a practical implementation for JSD. We conduct both theoretical and empirical studies to validate our method. Experimental results on T3Bench demonstrate that our method can produce high-quality and diversified 3D assets.

Uy Dieu Tran, Minh Luu, Phong Ha Nguyen et al.

Existing Score Distillation Sampling (SDS)-based methods have driven significant progress in text-to-3D generation. However, 3D models produced by SDS-based methods tend to exhibit over-smoothing and low-quality outputs. These issues arise from the mode-seeking behavior of current methods, where the scores used to update the model oscillate between multiple modes, resulting in unstable optimization and diminished output quality. To address this problem, we introduce a novel image prompt score distillation loss named ISD, which employs a reference image to direct text-to-3D optimization toward a specific mode. Our ISD loss can be implemented by using IP-Adapter, a lightweight adapter for integrating image prompt capability to a text-to-image diffusion model, as a mode-selection module. A variant of this adapter, when not being prompted by a reference image, can serve as an efficient control variate to reduce variance in score estimates, thereby enhancing both output quality and optimization stability. Our experiments demonstrate that the ISD loss consistently achieves visually coherent, high-quality outputs and improves optimization speed compared to prior text-to-3D methods, as demonstrated through both qualitative and quantitative evaluations on the T3Bench benchmark suite.

Tung Do, Thuan Hoang Nguyen, Anh Tuan Tran et al.

We propose a new view synthesis method via synthesizing a 3D neural field from both single or few-view input images. To address the ill-posed nature of the image-to-3D generation problem, we devise a two-stage method that involves a reconstruction model and a diffusion model for view synthesis. Our reconstruction model first lifts one or more input images to the 3D space from a volume as the coarse-scale 3D representation followed by a tri-plane as the fine-scale 3D representation. To mitigate the ambiguity in occluded regions, our diffusion model then hallucinates missing details in the rendered images from tri-planes. We then introduce a new progressive refinement technique that iteratively applies the reconstruction and diffusion model to gradually synthesize novel views, boosting the overall quality of the 3D representations and their rendering. Empirical evaluation demonstrates the superiority of our method over state-of-the-art methods on the synthetic SRN-Car dataset, the in-the-wild CO3D dataset, and large-scale Objaverse dataset while achieving both sampling efficacy and multi-view consistency.

Hubert Kompanowski, Binh-Son Hua

We present a method to generate 3D objects in styles. Our method takes a text prompt and a style reference image as input and reconstructs a neural radiance field to synthesize a 3D model with the content aligning with the text prompt and the style following the reference image. To simultaneously generate the 3D object and perform style transfer in one go, we propose a stylized score distillation loss to guide a text-to-3D optimization process to output visually plausible geometry and appearance. Our stylized score distillation is based on a combination of an original pretrained text-to-image model and its modified sibling with the key and value features of self-attention layers manipulated to inject styles from the reference image. Comparisons with state-of-the-art methods demonstrated the strong visual performance of our method, further supported by the quantitative results from our user study.

Quang-Trung Truong, Duc Thanh Nguyen, Binh-Son Hua et al.

Video object segmentation is a fundamental research problem in computer vision. Recent techniques have often applied attention mechanism to object representation learning from video sequences. However, due to temporal changes in the video data, attention maps may not well align with the objects of interest across video frames, causing accumulated errors in long-term video processing. In addition, existing techniques have utilised complex architectures, requiring highly computational complexity and hence limiting the ability to integrate video object segmentation into low-powered devices. To address these issues, we propose a new method for self-supervised video object segmentation based on distillation learning of deformable attention. Specifically, we devise a lightweight architecture for video object segmentation that is effectively adapted to temporal changes. This is enabled by deformable attention mechanism, where the keys and values capturing the memory of a video sequence in the attention module have flexible locations updated across frames. The learnt object representations are thus adaptive to both the spatial and temporal dimensions. We train the proposed architecture in a self-supervised fashion through a new knowledge distillation paradigm where deformable attention maps are integrated into the distillation loss. We qualitatively and quantitatively evaluate our method and compare it with existing methods on benchmark datasets including DAVIS 2016/2017 and YouTube-VOS 2018/2019. Experimental results verify the superiority of our method via its achieved state-of-the-art performance and optimal memory usage.

Zhiyuan Zhang, Binh-Son Hua, Sai-Kit Yeung

3D point clouds deep learning is a promising field of research that allows a neural network to learn features of point clouds directly, making it a robust tool for solving 3D scene understanding tasks. While recent works show that point cloud convolutions can be invariant to translation and point permutation, investigations of the rotation invariance property for point cloud convolution has been so far scarce. Some existing methods perform point cloud convolutions with rotation-invariant features, existing methods generally do not perform as well as translation-invariant only counterpart. In this work, we argue that a key reason is that compared to point coordinates, rotation-invariant features consumed by point cloud convolution are not as distinctive. To address this problem, we propose a simple yet effective convolution operator that enhances feature distinction by designing powerful rotation invariant features from the local regions. We consider the relationship between the point of interest and its neighbors as well as the internal relationship of the neighbors to largely improve the feature descriptiveness. Our network architecture can capture both local and global context by simply tuning the neighborhood size in each convolution layer. We conduct several experiments on synthetic and real-world point cloud classifications, part segmentation, and shape retrieval to evaluate our method, which achieves the state-of-the-art accuracy under challenging rotations.

Minh Tran, Loi Ly, Binh-Son Hua et al.

Capsule network is a recent new deep network architecture that has been applied successfully for medical image segmentation tasks. This work extends capsule networks for volumetric medical image segmentation with self-supervised learning. To improve on the problem of weight initialization compared to previous capsule networks, we leverage self-supervised learning for capsule networks pre-training, where our pretext-task is optimized by self-reconstruction. Our capsule network, SS-3DCapsNet, has a UNet-based architecture with a 3D Capsule encoder and 3D CNNs decoder. Our experiments on multiple datasets including iSeg-2017, Hippocampus, and Cardiac demonstrate that our 3D capsule network with self-supervised pre-training considerably outperforms previous capsule networks and 3D-UNets.

Tan M. Dinh, Rang Nguyen, Binh-Son Hua

In this paper, we conduct a study on the state-of-the-art methods for text-to-image synthesis and propose a framework to evaluate these methods. We consider syntheses where an image contains a single or multiple objects. Our study outlines several issues in the current evaluation pipeline: (i) for image quality assessment, a commonly used metric, e.g., Inception Score (IS), is often either miscalibrated for the single-object case or misused for the multi-object case; (ii) for text relevance and object accuracy assessment, there is an overfitting phenomenon in the existing R-precision (RP) and Semantic Object Accuracy (SOA) metrics, respectively; (iii) for multi-object case, many vital factors for evaluation, e.g., object fidelity, positional alignment, counting alignment, are largely dismissed; (iv) the ranking of the methods based on current metrics is highly inconsistent with real images. To overcome these issues, we propose a combined set of existing and new metrics to systematically evaluate the methods. For existing metrics, we offer an improved version of IS named IS* by using temperature scaling to calibrate the confidence of the classifier used by IS; we also propose a solution to mitigate the overfitting issues of RP and SOA. For new metrics, we develop counting alignment, positional alignment, object-centric IS, and object-centric FID metrics for evaluating the multi-object case. We show that benchmarking with our bag of metrics results in a highly consistent ranking among existing methods that is well-aligned with human evaluation. As a by-product, we create AttnGAN++, a simple but strong baseline for the benchmark by stabilizing the training of AttnGAN using spectral normalization. We also release our toolbox, so-called TISE, for advocating fair and consistent evaluation of text-to-image models.

Tan M. Dinh, Anh Tuan Tran, Rang Nguyen et al.

Real-world image manipulation has achieved fantastic progress in recent years as a result of the exploration and utilization of GAN latent spaces. GAN inversion is the first step in this pipeline, which aims to map the real image to the latent code faithfully. Unfortunately, the majority of existing GAN inversion methods fail to meet at least one of the three requirements listed below: high reconstruction quality, editability, and fast inference. We present a novel two-phase strategy in this research that fits all requirements at the same time. In the first phase, we train an encoder to map the input image to StyleGAN2 $\mathcal{W}$-space, which was proven to have excellent editability but lower reconstruction quality. In the second phase, we supplement the reconstruction ability in the initial phase by leveraging a series of hypernetworks to recover the missing information during inversion. These two steps complement each other to yield high reconstruction quality thanks to the hypernetwork branch and excellent editability due to the inversion done in the $\mathcal{W}$-space. Our method is entirely encoder-based, resulting in extremely fast inference. Extensive experiments on two challenging datasets demonstrate the superiority of our method.

Duong H. Le, Binh-Son Hua

Network pruning is an effective method to reduce the computational expense of over-parameterized neural networks for deployment on low-resource systems. Recent state-of-the-art techniques for retraining pruned networks such as weight rewinding and learning rate rewinding have been shown to outperform the traditional fine-tuning technique in recovering the lost accuracy (Renda et al., 2020), but so far it is unclear what accounts for such performance. In this work, we conduct extensive experiments to verify and analyze the uncanny effectiveness of learning rate rewinding. We find that the reason behind the success of learning rate rewinding is the usage of a large learning rate. Similar phenomenon can be observed in other learning rate schedules that involve large learning rates, e.g., the 1-cycle learning rate schedule (Smith et al., 2019). By leveraging the right learning rate schedule in retraining, we demonstrate a counter-intuitive phenomenon in that randomly pruned networks could even achieve better performance than methodically pruned networks (fine-tuned with the conventional approach). Our results emphasize the cruciality of the learning rate schedule in pruned network retraining - a detail often overlooked by practitioners during the implementation of network pruning. One-sentence Summary: We study the effective of different retraining mechanisms while doing pruning

Trung Nguyen, Quang-Hieu Pham, Tam Le et al.

Learning an effective representation of 3D point clouds requires a good metric to measure the discrepancy between two 3D point sets, which is non-trivial due to their irregularity. Most of the previous works resort to using the Chamfer discrepancy or Earth Mover's distance, but those metrics are either ineffective in measuring the differences between point clouds or computationally expensive. In this paper, we conduct a systematic study with extensive experiments on distance metrics for 3D point clouds. From this study, we propose to use sliced Wasserstein distance and its variants for learning representations of 3D point clouds. In addition, we introduce a new algorithm to estimate sliced Wasserstein distance that guarantees that the estimated value is close enough to the true one. Experiments show that the sliced Wasserstein distance and its variants allow the neural network to learn a more efficient representation compared to the Chamfer discrepancy. We demonstrate the efficiency of the sliced Wasserstein metric and its variants on several tasks in 3D computer vision including training a point cloud autoencoder, generative modeling, transfer learning, and point cloud registration.

Jaeyeon Kim, Binh-Son Hua, Duc Thanh Nguyen et al.

With recent developments of convolutional neural networks, deep learning for 3D point clouds has shown significant progress in various 3D scene understanding tasks, e.g., object recognition, semantic segmentation. In a safety-critical environment, it is however not well understood how such deep learning models are vulnerable to adversarial examples. In this work, we explore adversarial attacks for point cloud-based neural networks. We propose a unified formulation for adversarial point cloud generation that can generalise two different attack strategies. Our method generates adversarial examples by attacking the classification ability of point cloud-based networks while considering the perceptibility of the examples and ensuring the minimal level of point manipulations. Experimental results show that our method achieves the state-of-the-art performance with higher than 89% and 90% of attack success rate on synthetic and real-world data respectively, while manipulating only about 4% of the total points.

Zhiyuan Zhang, Binh-Son Hua, Wei Chen et al.

Recent advances in deep learning for 3D point clouds have shown great promises in scene understanding tasks thanks to the introduction of convolution operators to consume 3D point clouds directly in a neural network. Point cloud data, however, could have arbitrary rotations, especially those acquired from 3D scanning. Recent works show that it is possible to design point cloud convolutions with rotation invariance property, but such methods generally do not perform as well as translation-invariant only convolution. We found that a key reason is that compared to point coordinates, rotation-invariant features consumed by point cloud convolution are not as distinctive. To address this problem, we propose a novel convolution operator that enhances feature distinction by integrating global context information from the input point cloud to the convolution. To this end, a globally weighted local reference frame is constructed in each point neighborhood in which the local point set is decomposed into bins. Anchor points are generated in each bin to represent global shape features. A convolution can then be performed to transform the points and anchor features into final rotation-invariant features. We conduct several experiments on point cloud classification, part segmentation, shape retrieval, and normals estimation to evaluate our convolution, which achieves state-of-the-art accuracy under challenging rotations.

Quang-Hieu Pham, Mikaela Angelina Uy, Binh-Son Hua et al.

In this work, we present a novel method to learn a local cross-domain descriptor for 2D image and 3D point cloud matching. Our proposed method is a dual auto-encoder neural network that maps 2D and 3D input into a shared latent space representation. We show that such local cross-domain descriptors in the shared embedding are more discriminative than those obtained from individual training in 2D and 3D domains. To facilitate the training process, we built a new dataset by collecting $\approx 1.4$ millions of 2D-3D correspondences with various lighting conditions and settings from publicly available RGB-D scenes. Our descriptor is evaluated in three main experiments: 2D-3D matching, cross-domain retrieval, and sparse-to-dense depth estimation. Experimental results confirm the robustness of our approach as well as its competitive performance not only in solving cross-domain tasks but also in being able to generalize to solve sole 2D and 3D tasks. Our dataset and code are released publicly at \url{https://hkust-vgd.github.io/lcd}.

Zhiyuan Zhang, Binh-Son Hua, David W. Rosen et al.

Recent progresses in 3D deep learning has shown that it is possible to design special convolution operators to consume point cloud data. However, a typical drawback is that rotation invariance is often not guaranteed, resulting in networks being trained with data augmented with rotations. In this paper, we introduce a novel convolution operator for point clouds that achieves rotation invariance. Our core idea is to use low-level rotation invariant geometric features such as distances and angles to design a convolution operator for point cloud learning. The well-known point ordering problem is also addressed by a binning approach seamlessly built into the convolution. This convolution operator then serves as the basic building block of a neural network that is robust to point clouds under 6DoF transformations such as translation and rotation. Our experiment shows that our method performs with high accuracy in common scene understanding tasks such as object classification and segmentation. Compared to previous works, most importantly, our method is able to generalize and achieve consistent results across different scenarios in which training and testing can contain arbitrary rotations.

Zhiyuan Zhang, Binh-Son Hua, Sai-Kit Yeung

Deep learning with 3D data has progressed significantly since the introduction of convolutional neural networks that can handle point order ambiguity in point cloud data. While being able to achieve good accuracies in various scene understanding tasks, previous methods often have low training speed and complex network architecture. In this paper, we address these problems by proposing an efficient end-to-end permutation invariant convolution for point cloud deep learning. Our simple yet effective convolution operator named ShellConv uses statistics from concentric spherical shells to define representative features and resolve the point order ambiguity, allowing traditional convolution to perform on such features. Based on ShellConv we further build an efficient neural network named ShellNet to directly consume the point clouds with larger receptive fields while maintaining less layers. We demonstrate the efficacy of ShellNet by producing state-of-the-art results on object classification, object part segmentation, and semantic scene segmentation while keeping the network very fast to train.

Mikaela Angelina Uy, Quang-Hieu Pham, Binh-Son Hua et al.

Deep learning techniques for point cloud data have demonstrated great potentials in solving classical problems in 3D computer vision such as 3D object classification and segmentation. Several recent 3D object classification methods have reported state-of-the-art performance on CAD model datasets such as ModelNet40 with high accuracy (~92%). Despite such impressive results, in this paper, we argue that object classification is still a challenging task when objects are framed with real-world settings. To prove this, we introduce ScanObjectNN, a new real-world point cloud object dataset based on scanned indoor scene data. From our comprehensive benchmark, we show that our dataset poses great challenges to existing point cloud classification techniques as objects from real-world scans are often cluttered with background and/or are partial due to occlusions. We identify three key open problems for point cloud object classification, and propose new point cloud classification neural networks that achieve state-of-the-art performance on classifying objects with cluttered background. Our dataset and code are publicly available in our project page https://hkust-vgd.github.io/scanobjectnn/.

Quang-Hieu Pham, Duc Thanh Nguyen, Binh-Son Hua et al.

Deep learning techniques have become the to-go models for most vision-related tasks on 2D images. However, their power has not been fully realised on several tasks in 3D space, e.g., 3D scene understanding. In this work, we jointly address the problems of semantic and instance segmentation of 3D point clouds. Specifically, we develop a multi-task pointwise network that simultaneously performs two tasks: predicting the semantic classes of 3D points and embedding the points into high-dimensional vectors so that points of the same object instance are represented by similar embeddings. We then propose a multi-value conditional random field model to incorporate the semantic and instance labels and formulate the problem of semantic and instance segmentation as jointly optimising labels in the field model. The proposed method is thoroughly evaluated and compared with existing methods on different indoor scene datasets including S3DIS and SceneNN. Experimental results showed the robustness of the proposed joint semantic-instance segmentation scheme over its single components. Our method also achieved state-of-the-art performance on semantic segmentation.

Quang-Hieu Pham, Binh-Son Hua, Duc Thanh Nguyen et al.

The widespread adoption of autonomous systems such as drones and assistant robots has created a need for real-time high-quality semantic scene segmentation. In this paper, we propose an efficient yet robust technique for on-the-fly dense reconstruction and semantic segmentation of 3D indoor scenes. To guarantee (near) real-time performance, our method is built atop an efficient super-voxel clustering method and a conditional random field with higher-order constraints from structural and object cues, enabling progressive dense semantic segmentation without any precomputation. We extensively evaluate our method on different indoor scenes including kitchens, offices, and bedrooms in the SceneNN and ScanNet datasets and show that our technique consistently produces state-of-the-art segmentation results in both qualitative and quantitative experiments.

Binh-Son Hua, Minh-Khoi Tran, Sai-Kit Yeung

Deep learning with 3D data such as reconstructed point clouds and CAD models has received great research interests recently. However, the capability of using point clouds with convolutional neural network has been so far not fully explored. In this paper, we present a convolutional neural network for semantic segmentation and object recognition with 3D point clouds. At the core of our network is pointwise convolution, a new convolution operator that can be applied at each point of a point cloud. Our fully convolutional network design, while being surprisingly simple to implement, can yield competitive accuracy in both semantic segmentation and object recognition task.

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.

Duc Thanh Nguyen, Binh-Son Hua, Lap-Fai Yu et al.

Recent advances of 3D acquisition devices have enabled large-scale acquisition of 3D scene data. Such data, if completely and well annotated, can serve as useful ingredients for a wide spectrum of computer vision and graphics works such as data-driven modeling and scene understanding, object detection and recognition. However, annotating a vast amount of 3D scene data remains challenging due to the lack of an effective tool and/or the complexity of 3D scenes (e.g. clutter, varying illumination conditions). This paper aims to build a robust annotation tool that effectively and conveniently enables the segmentation and annotation of massive 3D data. Our tool works by coupling 2D and 3D information via an interactive framework, through which users can provide high-level semantic annotation for objects. We have experimented our tool and found that a typical indoor scene could be well segmented and annotated in less than 30 minutes by using the tool, as opposed to a few hours if done manually. Along with the tool, we created a dataset of over a hundred 3D scenes associated with complete annotations using our tool. The tool and dataset are available at www.scenenn.net.