Computer Vision and Pattern Recognition (cs.CV)

Abstract: Implicit Neural Representation (INR) is an innovative approach for representing complex shapes or objects without explicitly defining their geometry or surface structure. Instead, INR represents objects as continuous functions. Previous research has demonstrated the effectiveness of using neural networks as INR for image compression, showcasing comparable performance to traditional methods such as JPEG. However, INR holds potential for various applications beyond image compression. This paper introduces Rapid-INR, a novel approach that utilizes INR for encoding and compressing images, thereby accelerating neural network training in computer vision tasks. Our methodology involves storing the whole dataset directly in INR format on a GPU, mitigating the significant data communication overhead between the CPU and GPU during training. Additionally, the decoding process from INR to RGB format is highly parallelized and executed on-the-fly. To further enhance compression, we propose iterative and dynamic pruning, as well as layer-wise quantization, building upon previous work. We evaluate our framework on the image classification task, utilizing the ResNet-18 backbone network and three commonly used datasets with varying image sizes. Rapid-INR reduces memory consumption to only 5% of the original dataset size and achieves a maximum 6$\times$ speedup over the PyTorch training pipeline, as well as a maximum 1.2x speedup over the DALI training pipeline, with only a marginal decrease in accuracy. Importantly, Rapid-INR can be readily applied to other computer vision tasks and backbone networks with reasonable engineering efforts. Our implementation code is publicly available at https://anonymous.4open.science/r/INR-4BF7.

Abstract: This paper presents Diffusion Model for Scene Text Recognition (DiffusionSTR), an end-to-end text recognition framework using diffusion models for recognizing text in the wild. While existing studies have viewed the scene text recognition task as an image-to-text transformation, we rethought it as a text-text one under images in a diffusion model. We show for the first time that the diffusion model can be applied to text recognition. Furthermore, experimental results on publicly available datasets show that the proposed method achieves competitive accuracy compared to state-of-the-art methods.

Abstract: Long-tailed visual recognition has received increasing attention in recent years. Due to the extremely imbalanced data distribution in long-tailed learning, the learning process shows great uncertainties. For example, the predictions of different experts on the same image vary remarkably despite the same training settings. To alleviate the uncertainty, we propose a Nested Collaborative Learning (NCL++) which tackles the long-tailed learning problem by a collaborative learning. To be specific, the collaborative learning consists of two folds, namely inter-expert collaborative learning (InterCL) and intra-expert collaborative learning (IntraCL). In-terCL learns multiple experts collaboratively and concurrently, aiming to transfer the knowledge among different experts. IntraCL is similar to InterCL, but it aims to conduct the collaborative learning on multiple augmented copies of the same image within the single expert. To achieve the collaborative learning in long-tailed learning, the balanced online distillation is proposed to force the consistent predictions among different experts and augmented copies, which reduces the learning uncertainties. Moreover, in order to improve the meticulous distinguishing ability on the confusing categories, we further propose a Hard Category Mining (HCM), which selects the negative categories with high predicted scores as the hard categories. Then, the collaborative learning is formulated in a nested way, in which the learning is conducted on not just all categories from a full perspective but some hard categories from a partial perspective. Extensive experiments manifest the superiority of our method with outperforming the state-of-the-art whether with using a single model or an ensemble. The code will be publicly released.

Abstract: We study visual question answering in a setting where the answer has to be mined from a pool of relevant and irrelevant images given as a context. For such a setting, a model must first retrieve relevant images from the pool and answer the question from these retrieved images. We refer to this problem as retrieval-based visual question answering (or RETVQA in short). The RETVQA is distinctively different and more challenging than the traditionally-studied Visual Question Answering (VQA), where a given question has to be answered with a single relevant image in context. Towards solving the RETVQA task, we propose a unified Multi Image BART (MI-BART) that takes a question and retrieved images using our relevance encoder for free-form fluent answer generation. Further, we introduce the largest dataset in this space, namely RETVQA, which has the following salient features: multi-image and retrieval requirement for VQA, metadata-independent questions over a pool of heterogeneous images, expecting a mix of classification-oriented and open-ended generative answers. Our proposed framework achieves an accuracy of 76.5% and a fluency of 79.3% on the proposed dataset, namely RETVQA and also outperforms state-of-the-art methods by 4.9% and 11.8% on the image segment of the publicly available WebQA dataset on the accuracy and fluency metrics, respectively.

Abstract: Arbitrary-oriented object detection is a relatively emerging but challenging task. Although remarkable progress has been made, there still remain many unsolved issues due to the large diversity of patterns in orientation, scale, aspect ratio, and visual appearance of objects in aerial images. Most of the existing methods adopt a coarse-grained fixed label assignment strategy and suffer from the inconsistency between the classification score and localization accuracy. First, to align the metric inconsistency between sample selection and regression loss calculation caused by fixed IoU strategy, we introduce affine transformation to evaluate the quality of samples and propose a distance-based label assignment strategy. The proposed metric-aligned selection (MAS) strategy can dynamically select samples according to the shape and rotation characteristic of objects. Second, to further address the inconsistency between classification and localization, we propose a critical feature sampling (CFS) module, which performs localization refinement on the sampling location for classification task to extract critical features accurately. Third, we present a scale-controlled smooth $L_1$ loss (SC-Loss) to adaptively select high quality samples by changing the form of regression loss function based on the statistics of proposals during training. Extensive experiments are conducted on four challenging rotated object detection datasets DOTA, FAIR1M-1.0, HRSC2016, and UCAS-AOD. The results show the state-of-the-art accuracy of the proposed detector.

Abstract: The detection of leaf diseases in plants generally involves visual observation of patterns appearing on the leaf surface. However, there are many diseases that are distinguished based on very subtle changes in these visually observable patterns. This paper attempts to identify plant leaf diseases using image processing techniques. The focus of this study is on the detection of citrus leaf canker disease. Canker is a bacterial infection of leaves. Symptoms of citrus cankers include brown spots on the leaves, often with a watery or oily appearance. The spots (called lesions in botany) are usually yellow. It is surrounded by a halo of the leaves and is found on both the top and bottom of the leaf. This paper describes various methods that have been used to detect citrus leaf canker disease. The methods used are histogram comparison and k-means clustering. Using these methods, citrus canker development was detected based on histograms generated based on leaf patterns. The results thus obtained can be used, after consultation with experts in the field of agriculture, to identify suitable treatments for the processes used.

Abstract: Demystifying complex human-ground interactions is essential for accurate and realistic 3D human motion reconstruction from RGB videos, as it ensures consistency between the humans and the ground plane. Prior methods have modeled human-ground interactions either implicitly or in a sparse manner, often resulting in unrealistic and incorrect motions when faced with noise and uncertainty. In contrast, our approach explicitly represents these interactions in a dense and continuous manner. To this end, we propose a novel Ground-aware Motion Model for 3D Human Motion Reconstruction, named GraMMaR, which jointly learns the distribution of transitions in both pose and interaction between every joint and ground plane at each time step of a motion sequence. It is trained to explicitly promote consistency between the motion and distance change towards the ground. After training, we establish a joint optimization strategy that utilizes GraMMaR as a dual-prior, regularizing the optimization towards the space of plausible ground-aware motions. This leads to realistic and coherent motion reconstruction, irrespective of the assumed or learned ground plane. Through extensive evaluation on the AMASS and AIST++ datasets, our model demonstrates good generalization and discriminating abilities in challenging cases including complex and ambiguous human-ground interactions. The code will be released.

Abstract: Foundation models have exhibited remarkable success in various applications, such as disease diagnosis and text report generation. To date, a foundation model for endoscopic video analysis is still lacking. In this paper, we propose Endo-FM, a foundation model specifically developed using massive endoscopic video data. First, we build a video transformer, which captures both local and global long-range dependencies across spatial and temporal dimensions. Second, we pre-train our transformer model using global and local views via a self-supervised manner, aiming to make it robust to spatial-temporal variations and discriminative across different scenes. To develop the foundation model, we construct a large-scale endoscopy video dataset by combining 9 publicly available datasets and a privately collected dataset from Baoshan Branch of Renji Hospital in Shanghai, China. Our dataset overall consists of over 33K video clips with up to 5 million frames, encompassing various protocols, target organs, and disease types. Our pre-trained Endo-FM can be easily adopted for a given downtream task via fine-tuning by serving as the backbone. With experiments on 3 different types of downstream tasks, including classification, segmentation, and detection, our Endo-FM surpasses the current state-of-the-art self-supervised pre-training and adapter-based transfer learning methods by a significant margin, such as VCL (3.1% F1 for classification, 4.8% Dice for segmentation, and 5.5% F1 for detection) and ST-Adapter (5.9% F1 for classification, 9.6% Dice for segmentation, and 9.9% F1 for detection). Code, datasets, and models are released at https://github.com/med-air/Endo-FM.

Abstract: Hyperspectral images (HSI) captured from earth observing satellites and aircraft is becoming increasingly important for applications in agriculture, environmental monitoring, mining, etc. Due to the limited available hyperspectral datasets, the pixel-wise random sampling is the most commonly used training-test dataset partition approach, which has significant overlap between samples in training and test datasets. Furthermore, our experimental observations indicates that regions with larger overlap often exhibit higher classification accuracy. Consequently, the pixel-wise random sampling approach poses a risk of data leakage. Thus, we propose a block-wise sampling method to minimize the potential for data leakage. Our experimental findings also confirm the presence of data leakage in models such as 2DCNN. Further, We propose a spectral-spatial axial aggregation transformer model, namely SaaFormer, to address the challenges associated with hyperspectral image classifier that considers HSI as long sequential three-dimensional images. The model comprises two primary components: axial aggregation attention and multi-level spectral-spatial extraction. The axial aggregation attention mechanism effectively exploits the continuity and correlation among spectral bands at each pixel position in hyperspectral images, while aggregating spatial dimension features. This enables SaaFormer to maintain high precision even under block-wise sampling. The multi-level spectral-spatial extraction structure is designed to capture the sensitivity of different material components to specific spectral bands, allowing the model to focus on a broader range of spectral details. The results on six publicly available datasets demonstrate that our model exhibits comparable performance when using random sampling, while significantly outperforming other methods when employing block-wise sampling partition.

Abstract: The understanding of complex human interactions and group activities has garnered attention in human-centric computer vision. However, the advancement of the related tasks is hindered due to the difficulty of obtaining large-scale labeled real-world datasets. To mitigate the issue, we propose M3Act, a multi-view multi-group multi-person human atomic action and group activity data generator. Powered by the Unity engine, M3Act contains simulation-ready 3D scenes and human assets, configurable lighting and camera systems, highly parameterized modular group activities, and a large degree of domain randomization during the data generation process. Our data generator is capable of generating large-scale datasets of human activities with multiple viewpoints, modalities (RGB images, 2D poses, 3D motions), and high-quality annotations for individual persons and multi-person groups (2D bounding boxes, instance segmentation masks, individual actions and group activity categories). Using M3Act, we perform synthetic data pre-training for 2D skeleton-based group activity recognition and RGB-based multi-person pose tracking. The results indicate that learning from our synthetic datasets largely improves the model performances on real-world datasets, with the highest gain of 5.59% and 7.32% respectively in group and person recognition accuracy on CAD2, as well as an improvement of 6.63 in MOTP on HiEve. Pre-training with our synthetic data also leads to faster model convergence on downstream tasks (up to 6.8% faster). Moreover, M3Act opens new research problems for 3D group activity generation. We release M3Act3D, an 87.6-hour 3D motion dataset of human activities with larger group sizes and higher complexity of inter-person interactions than previous multi-person datasets. We define multiple metrics and propose a competitive baseline for the novel task.

Abstract: Decreased visibility, intensive noise, and biased color are the common problems existing in low-light images. These visual disturbances further reduce the performance of high-level vision tasks, such as object detection, and tracking. To address this issue, some image enhancement methods have been proposed to increase the image contrast. However, most of them are implemented only in the spatial domain, which can be severely influenced by noise signals while enhancing. Hence, in this work, we propose a novel residual recurrent multi-wavelet convolutional neural network R2-MWCNN learned in the frequency domain that can simultaneously increase the image contrast and reduce noise signals well. This end-to-end trainable network utilizes a multi-level discrete wavelet transform to divide input feature maps into distinct frequencies, resulting in a better denoise impact. A channel-wise loss function is proposed to correct the color distortion for more realistic results. Extensive experiments demonstrate that our proposed R2-MWCNN outperforms the state-of-the-art methods quantitively and qualitatively.

Abstract: The objective of augmented reality (AR) is to add digital content to natural images and videos to create an interactive experience between the user and the environment. Scene analysis and object recognition play a crucial role in AR, as they must be performed quickly and accurately. In this study, a new approach is proposed that involves using oriented bounding boxes with a detection and recognition deep network to improve performance and processing time. The approach is evaluated using two datasets: a real image dataset (DOTA dataset) commonly used for computer vision tasks, and a synthetic dataset that simulates different environmental, lighting, and acquisition conditions. The focus of the evaluation is on small objects, which are difficult to detect and recognise. The results indicate that the proposed approach tends to produce better Average Precision and greater accuracy for small objects in most of the tested conditions.

Abstract: Perceptually Aligned Gradients (PAG) refer to an intriguing property observed in robust image classification models, wherein their input gradients align with human perception and pose semantic meanings. While this phenomenon has gained significant research attention, it was solely studied in the context of unimodal vision-only architectures. In this work, we extend the study of PAG to Vision-Language architectures, which form the foundations for diverse image-text tasks and applications. Through an adversarial robustification finetuning of CLIP, we demonstrate that robust Vision-Language models exhibit PAG in contrast to their vanilla counterparts. This work reveals the merits of CLIP with PAG (CLIPAG) in several vision-language generative tasks. Notably, we show that seamlessly integrating CLIPAG in a "plug-n-play" manner leads to substantial improvements in vision-language generative applications. Furthermore, leveraging its PAG property, CLIPAG enables text-to-image generation without any generative model, which typically requires huge generators.

Abstract: The mainstream style transfer methods usually use pre-trained deep convolutional neural network (VGG) models as encoders, or use more complex model structures to achieve better style transfer effects. This leads to extremely slow processing speeds for practical tasks due to limited resources or higher resolution image processing, such as 4K images, severely hindering the practical application value of style transfer models. We introduce a lightweight and fast styletransfer model with controllable detail attention enhancement, named ICDaeLST. The model adopts a minimal, shallow, and small architecture, forming a very compact lightweight model for efficient forward inference. Although its structure is simple and has limited parameters, we achieve better overall color and texture structure matching by introducing a style discriminator, design additional global semantic invariance loss to preserve the semantic and structural information of the content image from a high-level global perspective, and design a shallow detail attention enhancement module to preserve the detail information of the content image from a low-level detail perspective. We also achieve controllable intensity during inference for the first time (adjusting the degree of detail retention and texture structure transfer based on subjective judgment) to meet different users' subjective evaluation of stylization effects. Compared with the current best-performing and most lightweight models, our model achieves better style transfer quality and better content structure and detail retention, while having a smaller model size (17-250 times smaller) and faster speed (0.26-6.5 times faster), and achieves the fastest processing speed of 0.38s on 4K high-resolution images.

Abstract: The Red Palm Weevil (RPW) is a highly destructive insect causing economic losses and impacting palm tree farming worldwide. This paper proposes an innovative approach for sustainable palm tree farming by utilizing advanced technologies for the early detection and management of RPW. Our approach combines computer vision, deep learning (DL), the Internet of Things (IoT), and geospatial data to detect and classify RPW-infested palm trees effectively. The main phases include; (1) DL classification using sound data from IoT devices, (2) palm tree detection using YOLOv8 on UAV images, and (3) RPW mapping using geospatial data. Our custom DL model achieves 100% precision and recall in detecting and localizing infested palm trees. Integrating geospatial data enables the creation of a comprehensive RPW distribution map for efficient monitoring and targeted management strategies. This technology-driven approach benefits agricultural authorities, farmers, and researchers in managing RPW infestations and safeguarding palm tree plantations' productivity.

Abstract: This paper proposes a multi-object tracking (MOT) algorithm for traffic monitoring using a drone equipped with optical and thermal cameras. Object detections on the images are obtained using a neural network for each type of camera. The cameras are modelled as direction-of-arrival (DOA) sensors. Each DOA detection follows a von-Mises Fisher distribution, whose mean direction is obtain by projecting a vehicle position on the ground to the camera. We then use the trajectory Poisson multi-Bernoulli mixture filter (TPMBM), which is a Bayesian MOT algorithm, to optimally estimate the set of vehicle trajectories. We have also developed a parameter estimation algorithm for the measurement model. We have tested the accuracy of the resulting TPMBM filter in synthetic and experimental data sets.

Abstract: Estimating camera motion in deformable scenes poses a complex and open research challenge. Most existing non-rigid structure from motion techniques assume to observe also static scene parts besides deforming scene parts in order to establish an anchoring reference. However, this assumption does not hold true in certain relevant application cases such as endoscopies. Deformable odometry and SLAM pipelines, which tackle the most challenging scenario of exploratory trajectories, suffer from a lack of robustness and proper quantitative evaluation methodologies. To tackle this issue with a common benchmark, we introduce the Drunkard's Dataset, a challenging collection of synthetic data targeting visual navigation and reconstruction in deformable environments. This dataset is the first large set of exploratory camera trajectories with ground truth inside 3D scenes where every surface exhibits non-rigid deformations over time. Simulations in realistic 3D buildings lets us obtain a vast amount of data and ground truth labels, including camera poses, RGB images and depth, optical flow and normal maps at high resolution and quality. We further present a novel deformable odometry method, dubbed the Drunkard's Odometry, which decomposes optical flow estimates into rigid-body camera motion and non-rigid scene deformations. In order to validate our data, our work contains an evaluation of several baselines as well as a novel tracking error metric which does not require ground truth data. Dataset and code: https://davidrecasens.github.io/TheDrunkard'sOdometry/

Abstract: Pretraining with large-scale 3D volumes has a potential for improving the segmentation performance on a target medical image dataset where the training images and annotations are limited. Due to the high cost of acquiring pixel-level segmentation annotations on the large-scale pretraining dataset, pretraining with unannotated images is highly desirable. In this work, we propose a novel self-supervised learning strategy named Volume Fusion (VF) for pretraining 3D segmentation models. It fuses several random patches from a foreground sub-volume to a background sub-volume based on a predefined set of discrete fusion coefficients, and forces the model to predict the fusion coefficient of each voxel, which is formulated as a self-supervised segmentation task without manual annotations. Additionally, we propose a novel network architecture based on parallel convolution and transformer blocks that is suitable to be transferred to different downstream segmentation tasks with various scales of organs and lesions. The proposed model was pretrained with 110k unannotated 3D CT volumes, and experiments with different downstream segmentation targets including head and neck organs, thoracic/abdominal organs showed that our pretrained model largely outperformed training from scratch and several state-of-the-art self-supervised training methods and segmentation models. The code and pretrained model are available at https://github.com/openmedlab/MIS-FM.

Abstract: Single image 3D reconstruction is an important but challenging task that requires extensive knowledge of our natural world. Many existing methods solve this problem by optimizing a neural radiance field under the guidance of 2D diffusion models but suffer from lengthy optimization time, 3D inconsistency results, and poor geometry. In this work, we propose a novel method that takes a single image of any object as input and generates a full 360-degree 3D textured mesh in a single feed-forward pass. Given a single image, we first use a view-conditioned 2D diffusion model, Zero123, to generate multi-view images for the input view, and then aim to lift them up to 3D space. Since traditional reconstruction methods struggle with inconsistent multi-view predictions, we build our 3D reconstruction module upon an SDF-based generalizable neural surface reconstruction method and propose several critical training strategies to enable the reconstruction of 360-degree meshes. Without costly optimizations, our method reconstructs 3D shapes in significantly less time than existing methods. Moreover, our method favors better geometry, generates more 3D consistent results, and adheres more closely to the input image. We evaluate our approach on both synthetic data and in-the-wild images and demonstrate its superiority in terms of both mesh quality and runtime. In addition, our approach can seamlessly support the text-to-3D task by integrating with off-the-shelf text-to-image diffusion models.

Abstract: This paper introduces DreamDiffusion, a novel method for generating high-quality images directly from brain electroencephalogram (EEG) signals, without the need to translate thoughts into text. DreamDiffusion leverages pre-trained text-to-image models and employs temporal masked signal modeling to pre-train the EEG encoder for effective and robust EEG representations. Additionally, the method further leverages the CLIP image encoder to provide extra supervision to better align EEG, text, and image embeddings with limited EEG-image pairs. Overall, the proposed method overcomes the challenges of using EEG signals for image generation, such as noise, limited information, and individual differences, and achieves promising results. Quantitative and qualitative results demonstrate the effectiveness of the proposed method as a significant step towards portable and low-cost ``thoughts-to-image'', with potential applications in neuroscience and computer vision.

Abstract: We show, for the first time, that neural networks trained only on synthetic data achieve state-of-the-art accuracy on the problem of 3D human pose and shape (HPS) estimation from real images. Previous synthetic datasets have been small, unrealistic, or lacked realistic clothing. Achieving sufficient realism is non-trivial and we show how to do this for full bodies in motion. Specifically, our BEDLAM dataset contains monocular RGB videos with ground-truth 3D bodies in SMPL-X format. It includes a diversity of body shapes, motions, skin tones, hair, and clothing. The clothing is realistically simulated on the moving bodies using commercial clothing physics simulation. We render varying numbers of people in realistic scenes with varied lighting and camera motions. We then train various HPS regressors using BEDLAM and achieve state-of-the-art accuracy on real-image benchmarks despite training with synthetic data. We use BEDLAM to gain insights into what model design choices are important for accuracy. With good synthetic training data, we find that a basic method like HMR approaches the accuracy of the current SOTA method (CLIFF). BEDLAM is useful for a variety of tasks and all images, ground truth bodies, 3D clothing, support code, and more are available for research purposes. Additionally, we provide detailed information about our synthetic data generation pipeline, enabling others to generate their own datasets. See the project page: https://bedlam.is.tue.mpg.de/.

Abstract: Feature alignment is the primary means of fusing multimodal data. We propose a feature alignment method that fully fuses multimodal information, which alternately shifts and expands feature information from different modalities to have a consistent representation in a feature space. The proposed method can robustly capture high-level interactions between features of different modalities, thus significantly improving the performance of multimodal learning. We also show that the proposed method outperforms other popular multimodal schemes on multiple tasks. Experimental evaluation of ETT and MIT-BIH-Arrhythmia, datasets shows that the proposed method achieves state of the art performance.

Abstract: One central challenge in source-free unsupervised domain adaptation (UDA) is the lack of an effective approach to evaluate the prediction results of the adapted network model in the target domain. To address this challenge, we propose to explore a new method called cross-inferential networks (CIN). Our main idea is that, when we adapt the network model to predict the sample labels from encoded features, we use these prediction results to construct new training samples with derived labels to learn a new examiner network that performs a different but compatible task in the target domain. Specifically, in this work, the base network model is performing image classification while the examiner network is tasked to perform relative ordering of triplets of samples whose training labels are carefully constructed from the prediction results of the base network model. Two similarity measures, cross-network correlation matrix similarity and attention consistency, are then developed to provide important guidance for the UDA process. Our experimental results on benchmark datasets demonstrate that our proposed CIN approach can significantly improve the performance of source-free UDA.

Abstract: Classifiers based on deep neural networks have been recently challenged by Adversarial Attack, where the widely existing vulnerability has invoked the research in defending them from potential threats. Given a vulnerable classifier, existing defense methods are mostly white-box and often require re-training the victim under modified loss functions/training regimes. While the model/data/training specifics of the victim are usually unavailable to the user, re-training is unappealing, if not impossible for reasons such as limited computational resources. To this end, we propose a new black-box defense framework. It can turn any pre-trained classifier into a resilient one with little knowledge of the model specifics. This is achieved by new joint Bayesian treatments on the clean data, the adversarial examples and the classifier, for maximizing their joint probability. It is further equipped with a new post-train strategy which keeps the victim intact. We name our framework Bayesian Boundary Correction (BBC). BBC is a general and flexible framework that can easily adapt to different data types. We instantiate BBC for image classification and skeleton-based human activity recognition, for both static and dynamic data. Exhaustive evaluation shows that BBC has superior robustness and can enhance robustness without severely hurting the clean accuracy, compared with existing defense methods.

Abstract: Multimodal Named Entity Recognition (MNER) is a crucial task for information extraction from social media platforms such as Twitter. Most current methods rely on attention weights to extract information from both text and images but are often unreliable and lack interpretability. To address this problem, we propose incorporating uncertainty estimation into the MNER task, producing trustworthy predictions. Our proposed algorithm models the distribution of each modality as a Normal-inverse Gamma distribution, and fuses them into a unified distribution with an evidential fusion mechanism, enabling hierarchical characterization of uncertainties and promotion of prediction accuracy and trustworthiness. Additionally, we explore the potential of pre-trained large foundation models in MNER and propose an efficient fusion approach that leverages their robust feature representations. Experiments on two datasets demonstrate that our proposed method outperforms the baselines and achieves new state-of-the-art performance.

Abstract: State-of-the-art deep learning-based registration methods employ three different learning strategies: supervised learning, which requires costly manual annotations, unsupervised learning, which heavily relies on hand-crafted similarity metrics designed by domain experts, or learning from synthetic data, which introduces a domain shift. To overcome the limitations of these strategies, we propose a novel self-supervised learning paradigm for unsupervised registration, relying on self-training. Our idea is based on two key insights. Feature-based differentiable optimizers 1) perform reasonable registration even from random features and 2) stabilize the training of the preceding feature extraction network on noisy labels. Consequently, we propose cyclical self-training, where pseudo labels are initialized as the displacement fields inferred from random features and cyclically updated based on more and more expressive features from the learning feature extractor, yielding a self-reinforcement effect. We evaluate the method for abdomen and lung registration, consistently surpassing metric-based supervision and outperforming diverse state-of-the-art competitors. Source code is available at https://github.com/multimodallearning/reg-cyclical-self-train.

Abstract: Regularization is a set of techniques that are used to improve the generalization ability of deep neural networks. In this paper, we introduce spectral batch normalization (SBN), a novel effective method to improve generalization by normalizing feature maps in the frequency (spectral) domain. The activations of residual networks without batch normalization (BN) tend to explode exponentially in the depth of the network at initialization. This leads to extremely large feature map norms even though the parameters are relatively small. These explosive dynamics can be very detrimental to learning. BN makes weight decay regularization on the scaling factors $\gamma, \beta$ approximately equivalent to an additive penalty on the norm of the feature maps, which prevents extremely large feature map norms to a certain degree. However, we show experimentally that, despite the approximate additive penalty of BN, feature maps in deep neural networks (DNNs) tend to explode at the beginning of the network and that feature maps of DNNs contain large values during the whole training. This phenomenon also occurs in a weakened form in non-residual networks. SBN addresses large feature maps by normalizing them in the frequency domain. In our experiments, we empirically show that SBN prevents exploding feature maps at initialization and large feature map values during the training. Moreover, the normalization of feature maps in the frequency domain leads to more uniform distributed frequency components. This discourages the DNNs to rely on single frequency components of feature maps. These, together with other effects of SBN, have a regularizing effect on the training of residual and non-residual networks. We show experimentally that using SBN in addition to standard regularization methods improves the performance of DNNs by a relevant margin, e.g. ResNet50 on ImageNet by 0.71%.

Abstract: Multiple Object Tracking (MOT) is crucial to autonomous vehicle perception. End-to-end transformer-based algorithms, which detect and track objects simultaneously, show great potential for the MOT task. However, most existing methods focus on image-based tracking with a single object category. In this paper, we propose an end-to-end transformer-based MOT algorithm (MotionTrack) with multi-modality sensor inputs to track objects with multiple classes. Our objective is to establish a transformer baseline for the MOT in an autonomous driving environment. The proposed algorithm consists of a transformer-based data association (DA) module and a transformer-based query enhancement module to achieve MOT and Multiple Object Detection (MOD) simultaneously. The MotionTrack and its variations achieve better results (AMOTA score at 0.55) on the nuScenes dataset compared with other classical baseline models, such as the AB3DMOT, the CenterTrack, and the probabilistic 3D Kalman filter. In addition, we prove that a modified attention mechanism can be utilized for DA to accomplish the MOT, and aggregate history features to enhance the MOD performance.

Abstract: Approaching the era of ubiquitous computing, human motion sensing plays a crucial role in smart systems for decision making, user interaction, and personalized services. Extensive research has been conducted on human tracking, pose estimation, gesture recognition, and activity recognition, which are predominantly based on cameras in traditional methods. However, the intrusive nature of cameras limits their use in smart home applications. To address this, mmWave radars have gained popularity due to their privacy-friendly features. In this work, we propose \textit{milliFlow}, a novel deep learning method for scene flow estimation as a complementary motion information for mmWave point cloud, serving as an intermediate level of features and directly benefiting downstream human motion sensing tasks. Experimental results demonstrate the superior performance of our method with an average 3D endpoint error of 4.6cm, significantly surpassing the competing approaches. Furthermore, by incorporating scene flow information, we achieve remarkable improvements in human activity recognition, human parsing, and human body part tracking. To foster further research in this area, we provide our codebase and dataset for open access.

Abstract: One of the mainstream schemes for 2D human pose estimation (HPE) is learning keypoints heatmaps by a neural network. Existing methods typically improve the quality of heatmaps by customized architectures, such as high-resolution representation and vision Transformers. In this paper, we propose \textbf{DiffusionPose}, a new scheme that formulates 2D HPE as a keypoints heatmaps generation problem from noised heatmaps. During training, the keypoints are diffused to random distribution by adding noises and the diffusion model learns to recover ground-truth heatmaps from noised heatmaps with respect to conditions constructed by image feature. During inference, the diffusion model generates heatmaps from initialized heatmaps in a progressive denoising way. Moreover, we further explore improving the performance of DiffusionPose with conditions from human structural information. Extensive experiments show the prowess of our DiffusionPose, with improvements of 1.6, 1.2, and 1.2 mAP on widely-used COCO, CrowdPose, and AI Challenge datasets, respectively.

Abstract: The rapid advancements in computer vision have stimulated remarkable progress in face forgery techniques, capturing the dedicated attention of researchers committed to detecting forgeries and precisely localizing manipulated areas. Nonetheless, with limited fine-grained pixel-wise supervision labels, deepfake detection models perform unsatisfactorily on precise forgery detection and localization. To address this challenge, we introduce the well-trained vision segmentation foundation model, i.e., Segment Anything Model (SAM) in face forgery detection and localization. Based on SAM, we propose the Detect Any Deepfakes (DADF) framework with the Multiscale Adapter, which can capture short- and long-range forgery contexts for efficient fine-tuning. Moreover, to better identify forged traces and augment the model's sensitivity towards forgery regions, Reconstruction Guided Attention (RGA) module is proposed. The proposed framework seamlessly integrates end-to-end forgery localization and detection optimization. Extensive experiments on three benchmark datasets demonstrate the superiority of our approach for both forgery detection and localization. The codes will be released soon at https://github.com/laiyingxin2/DADF.

Abstract: Continual learning (CL) is an approach to address catastrophic forgetting, which refers to forgetting previously learned knowledge by neural networks when trained on new tasks or data distributions. The adversarial robustness has decomposed features into robust and non-robust types and demonstrated that models trained on robust features significantly enhance adversarial robustness. However, no study has been conducted on the efficacy of robust features from the lens of the CL model in mitigating catastrophic forgetting in CL. In this paper, we introduce the CL robust dataset and train four baseline models on both the standard and CL robust datasets. Our results demonstrate that the CL models trained on the CL robust dataset experienced less catastrophic forgetting of the previously learned tasks than when trained on the standard dataset. Our observations highlight the significance of the features provided to the underlying CL models, showing that CL robust features can alleviate catastrophic forgetting.

Abstract: Historically, the rotorcraft community has experienced a higher fatal accident rate than other aviation segments, including commercial and general aviation. Recent advancements in artificial intelligence (AI) and the application of these technologies in different areas of our lives are both intriguing and encouraging. When developed appropriately for the aviation domain, AI techniques provide an opportunity to help design systems that can address rotorcraft safety challenges. Our recent work demonstrated that AI algorithms could use video data from onboard cameras and correctly identify different flight parameters from cockpit gauges, e.g., indicated airspeed. These AI-based techniques provide a potentially cost-effective solution, especially for small helicopter operators, to record the flight state information and perform post-flight analyses. We also showed that carefully designed and trained AI systems could accurately predict rotorcraft attitude (i.e., pitch and yaw) from outside scenes (images or video data). Ordinary off-the-shelf video cameras were installed inside the rotorcraft cockpit to record the outside scene, including the horizon. The AI algorithm could correctly identify rotorcraft attitude at an accuracy in the range of 80\%. In this work, we combined five different onboard camera viewpoints to improve attitude prediction accuracy to 94\%. In this paper, five onboard camera views included the pilot windshield, co-pilot windshield, pilot Electronic Flight Instrument System (EFIS) display, co-pilot EFIS display, and the attitude indicator gauge. Using video data from each camera view, we trained various convolutional neural networks (CNNs), which achieved prediction accuracy in the range of 79\% % to 90\% %. We subsequently ensembled the learned knowledge from all CNNs and achieved an ensembled accuracy of 93.3\%.

Abstract: Instruction tuning unlocks the superior capability of Large Language Models (LLM) to interact with humans. Furthermore, recent instruction-following datasets include images as visual inputs, collecting responses for image-based instructions. However, visual instruction-tuned models cannot comprehend textual details within images well. This work enhances the current visual instruction tuning pipeline with text-rich images (e.g., movie posters, book covers, etc.). Specifically, we first use publicly available OCR tools to collect results on 422K text-rich images from the LAION dataset. Moreover, we prompt text-only GPT-4 with recognized texts and image captions to generate 16K conversations, each containing question-answer pairs for text-rich images. By combining our collected data with previous multi-modal instruction-following data, our model, LLaVAR, substantially improves the LLaVA model's capability on text-based VQA datasets (up to 20% accuracy improvement) while achieving an accuracy of 91.42% on ScienceQA. The GPT-4-based instruction-following evaluation also demonstrates the improvement of our model on both natural images and text-rich images. Through qualitative analysis, LLaVAR shows promising interaction (e.g., reasoning, writing, and elaboration) skills with humans based on the latest real-world online content that combines text and images. We make our code/data/models publicly available at https://llavar.github.io/.

Abstract: We present a novel alignment-before-generation approach to tackle the challenging task of generating general 3D shapes based on 2D images or texts. Directly learning a conditional generative model from images or texts to 3D shapes is prone to producing inconsistent results with the conditions because 3D shapes have an additional dimension whose distribution significantly differs from that of 2D images and texts. To bridge the domain gap among the three modalities and facilitate multi-modal-conditioned 3D shape generation, we explore representing 3D shapes in a shape-image-text-aligned space. Our framework comprises two models: a Shape-Image-Text-Aligned Variational Auto-Encoder (SITA-VAE) and a conditional Aligned Shape Latent Diffusion Model (ASLDM). The former model encodes the 3D shapes into the shape latent space aligned to the image and text and reconstructs the fine-grained 3D neural fields corresponding to given shape embeddings via the transformer-based decoder. The latter model learns a probabilistic mapping function from the image or text space to the latent shape space. Our extensive experiments demonstrate that our proposed approach can generate higher-quality and more diverse 3D shapes that better semantically conform to the visual or textural conditional inputs, validating the effectiveness of the shape-image-text-aligned space for cross-modality 3D shape generation.

Abstract: Masked image modelling (MIM) is a powerful self-supervised representation learning paradigm, whose potential has not been widely demonstrated in medical image analysis. In this work, we show the capacity of MIM to capture rich semantic representations of Haemotoxylin & Eosin (H&E)-stained images at the nuclear level. Inspired by Bidirectional Encoder representation from Image Transformers (BEiT), we split the images into smaller patches and generate corresponding discrete visual tokens. In addition to the regular grid-based patches, typically used in visual Transformers, we introduce patches of individual cell nuclei. We propose positional encoding of the irregular distribution of these structures within an image. We pre-train the model in a self-supervised manner on H&E-stained whole-slide images of diffuse large B-cell lymphoma, where cell nuclei have been segmented. The pre-training objective is to recover the original discrete visual tokens of the masked image on the one hand, and to reconstruct the visual tokens of the masked object instances on the other. Coupling these two pre-training tasks allows us to build powerful, context-aware representations of nuclei. Our model generalizes well and can be fine-tuned on downstream classification tasks, achieving improved cell classification accuracy on PanNuke dataset by more than 5% compared to current instance segmentation methods.

Abstract: Portrait synthesis creates realistic digital avatars which enable users to interact with others in a compelling way. Recent advances in StyleGAN and its extensions have shown promising results in synthesizing photorealistic and accurate reconstruction of human faces. However, previous methods often focus on frontal face synthesis and most methods are not able to handle large head rotations due to the training data distribution of StyleGAN. In this work, our goal is to take as input a monocular video of a face, and create an editable dynamic portrait able to handle extreme head poses. The user can create novel viewpoints, edit the appearance, and animate the face. Our method utilizes pivotal tuning inversion (PTI) to learn a personalized video prior from a monocular video sequence. Then we can input pose and expression coefficients to MLPs and manipulate the latent vectors to synthesize different viewpoints and expressions of the subject. We also propose novel loss functions to further disentangle pose and expression in the latent space. Our algorithm shows much better performance over previous approaches on monocular video datasets, and it is also capable of running in real-time at 54 FPS on an RTX 3080.

Abstract: Given sparse views of an object, estimating their camera poses is a long-standing and intractable problem. We harness the pre-trained diffusion model of novel views conditioned on viewpoints (Zero-1-to-3). We present ID-Pose which inverses the denoising diffusion process to estimate the relative pose given two input images. ID-Pose adds a noise on one image, and predicts the noise conditioned on the other image and a decision variable for the pose. The prediction error is used as the objective to find the optimal pose with the gradient descent method. ID-Pose can handle more than two images and estimate each of the poses with multiple image pairs from triangular relationships. ID-Pose requires no training and generalizes to real-world images. We conduct experiments using high-quality real-scanned 3D objects, where ID-Pose significantly outperforms state-of-the-art methods.

Abstract: Recent advances in diffusion-based generative models have shown incredible promise for Image-to-Image translation and editing. Most recent work in this space relies on additional training or architecture-specific adjustments to the diffusion process. In this work, we show that much of this low-level control can be achieved without additional training or any access to features of the diffusion model. Our method simply applies a filter to the input of each diffusion step based on the output of the previous step in an adaptive manner. Notably, this approach does not depend on any specific architecture or sampler and can be done without access to internal features of the network, making it easy to combine with other techniques, samplers, and diffusion architectures. Furthermore, it has negligible cost to performance, and allows for more continuous adjustment of guidance strength than other approaches. We show FGD offers a fast and strong baseline that is competitive with recent architecture-dependent approaches. Furthermore, FGD can also be used as a simple add-on to enhance the structural guidance of other state-of-the-art I2I methods. Finally, our derivation of this method helps to understand the impact of self attention, a key component of other recent architecture-specific I2I approaches, in a more architecture-independent way. Project page: https://github.com/jaclyngu/FilteredGuidedDiffusion

Abstract: Text-to-image diffusion models have attracted considerable interest due to their wide applicability across diverse fields. However, challenges persist in creating controllable models for personalized object generation. In this paper, we first identify the entanglement issues in existing personalized generative models, and then propose a straightforward and efficient data augmentation training strategy that guides the diffusion model to focus solely on object identity. By inserting the plug-and-play adapter layers from a pre-trained controllable diffusion model, our model obtains the ability to control the location and size of each generated personalized object. During inference, we propose a regionally-guided sampling technique to maintain the quality and fidelity of the generated images. Our method achieves comparable or superior fidelity for personalized objects, yielding a robust, versatile, and controllable text-to-image diffusion model that is capable of generating realistic and personalized images. Our approach demonstrates significant potential for various applications, such as those in art, entertainment, and advertising design.

Abstract: We present a model that can perform multiple vision tasks and can be adapted to other downstream tasks efficiently. Despite considerable progress in multi-task learning, most efforts focus on learning from multi-label data: a single image set with multiple task labels. Such multi-label data sets are rare, small, and expensive. We say heterogeneous to refer to image sets with different task labels, or to combinations of single-task datasets. Few have explored training on such heterogeneous datasets. General-purpose vision models are still dominated by single-task pretraining, and it remains unclear how to scale up multi-task models by leveraging mainstream vision datasets designed for different purposes. The challenges lie in managing large intrinsic differences among vision tasks, including data distribution, architectures, task-specific modules, dataset scales, and sampling strategies. To address these challenges, we propose to modify and scale up mixture-of-experts (MoE) vision transformers, so that they can simultaneously learn classification, detection, and segmentation on diverse mainstream vision datasets including ImageNet, COCO, and ADE20K. Our approach achieves comparable results to single-task state-of-the-art models and demonstrates strong generalization on downstream tasks. Due to its emergent modularity, this general-purpose model decomposes into high-performing components, efficiently adapting to downstream tasks. We can fine-tune it with fewer training parameters, fewer model parameters, and less computation. Additionally, its modularity allows for easy expansion in continual-learning-without-forgetting scenarios. Finally, these functions can be controlled and combined to meet various demands of downstream tasks.