Computer Vision and Pattern Recognition (cs.CV)

Abstract: Knowledge Graph (KG) plays a crucial role in Medical Report Generation (MRG) because it reveals the relations among diseases and thus can be utilized to guide the generation process. However, constructing a comprehensive KG is labor-intensive and its applications on the MRG process are under-explored. In this study, we establish a complete KG on chest X-ray imaging that includes 137 types of diseases and abnormalities. Based on this KG, we find that the current MRG data sets exhibit a long-tailed problem in disease distribution. To mitigate this problem, we introduce a novel augmentation strategy that enhances the representation of disease types in the tail-end of the distribution. We further design a two-stage MRG approach, where a classifier is first trained to detect whether the input images exhibit any abnormalities. The classified images are then independently fed into two transformer-based generators, namely, ``disease-specific generator" and ``disease-free generator" to generate the corresponding reports. To enhance the clinical evaluation of whether the generated reports correctly describe the diseases appearing in the input image, we propose diverse sensitivity (DS), a new metric that checks whether generated diseases match ground truth and measures the diversity of all generated diseases. Results show that the proposed two-stage generation framework and augmentation strategies improve DS by a considerable margin, indicating a notable reduction in the long-tailed problem associated with under-represented diseases.

Abstract: Pre-training has been widely adopted in deep learning to improve model performance, especially when the training data for a target task is limited. In our work, we seek to understand the implications of this training strategy on the generalization properties of downstream models. More specifically, we ask the following question: how do properties of the pre-training distribution affect the robustness of a fine-tuned model? The properties we explore include the label space, label semantics, image diversity, data domains, and data quantity of the pre-training distribution. We find that the primary factor influencing downstream effective robustness (Taori et al., 2020) is data quantity, while other factors have limited significance. For example, reducing the number of ImageNet pre-training classes by 4x while increasing the number of images per class by 4x (that is, keeping total data quantity fixed) does not impact the robustness of fine-tuned models. We demonstrate our findings on pre-training distributions drawn from various natural and synthetic data sources, primarily using the iWildCam-WILDS distribution shift as a test for downstream robustness.

Abstract: The existing deepfake detection methods have reached a bottleneck in generalizing to unseen forgeries and manipulation approaches. Based on the observation that the deepfake detectors exhibit a preference for overfitting the specific primary regions in input, this paper enhances the generalization capability from a novel regularization perspective. This can be simply achieved by augmenting the images through primary region removal, thereby preventing the detector from over-relying on data bias. Our method consists of two stages, namely the static localization for primary region maps, as well as the dynamic exploitation of primary region masks. The proposed method can be seamlessly integrated into different backbones without affecting their inference efficiency. We conduct extensive experiments over three widely used deepfake datasets - DFDC, DF-1.0, and Celeb-DF with five backbones. Our method demonstrates an average performance improvement of 6% across different backbones and performs competitively with several state-of-the-art baselines.

Abstract: The core idea of visual anomaly detection is to learn the normality from normal images, but previous works have been developed specifically for certain tasks, leading to fragmentation among various tasks: defect detection, semantic anomaly detection, multi-class anomaly detection, and anomaly clustering. This one-task-one-model approach is resource-intensive and incurs high maintenance costs as the number of tasks increases. This paper presents SelFormaly, a universal and powerful anomaly detection framework. We emphasize the necessity of our off-the-shelf approach by pointing out a suboptimal issue with fluctuating performance in previous online encoder-based methods. In addition, we question the effectiveness of using ConvNets as previously employed in the literature and confirm that self-supervised ViTs are suitable for unified anomaly detection. We introduce back-patch masking and discover the new role of top k-ratio feature matching to achieve unified and powerful anomaly detection. Back-patch masking eliminates irrelevant regions that possibly hinder target-centric detection with representations of the scene layout. The top k-ratio feature matching unifies various anomaly levels and tasks. Finally, SelFormaly achieves state-of-the-art results across various datasets for all the aforementioned tasks.

Abstract: Despite recent progress in enhancing the privacy of federated learning (FL) via differential privacy (DP), the trade-off of DP between privacy protection and performance is still underexplored for real-world medical scenario. In this paper, we propose to optimize the trade-off under the context of client-level DP, which focuses on privacy during communications. However, FL for medical imaging involves typically much fewer participants (hospitals) than other domains (e.g., mobile devices), thus ensuring clients be differentially private is much more challenging. To tackle this problem, we propose an adaptive intermediary strategy to improve performance without harming privacy. Specifically, we theoretically find splitting clients into sub-clients, which serve as intermediaries between hospitals and the server, can mitigate the noises introduced by DP without harming privacy. Our proposed approach is empirically evaluated on both classification and segmentation tasks using two public datasets, and its effectiveness is demonstrated with significant performance improvements and comprehensive analytical studies. Code is available at: https://github.com/med-air/Client-DP-FL.

Abstract: Video anomaly detection (VAD) has been paid increasing attention due to its potential applications, its current dominant tasks focus on online detecting anomalies% at the frame level, which can be roughly interpreted as the binary or multiple event classification. However, such a setup that builds relationships between complicated anomalous events and single labels, e.g., ``vandalism'', is superficial, since single labels are deficient to characterize anomalous events. In reality, users tend to search a specific video rather than a series of approximate videos. Therefore, retrieving anomalous events using detailed descriptions is practical and positive but few researches focus on this. In this context, we propose a novel task called Video Anomaly Retrieval (VAR), which aims to pragmatically retrieve relevant anomalous videos by cross-modalities, e.g., language descriptions and synchronous audios. Unlike the current video retrieval where videos are assumed to be temporally well-trimmed with short duration, VAR is devised to retrieve long untrimmed videos which may be partially relevant to the given query. To achieve this, we present two large-scale VAR benchmarks, UCFCrime-AR and XDViolence-AR, constructed on top of prevalent anomaly datasets. Meanwhile, we design a model called Anomaly-Led Alignment Network (ALAN) for VAR. In ALAN, we propose an anomaly-led sampling to focus on key segments in long untrimmed videos. Then, we introduce an efficient pretext task to enhance semantic associations between video-text fine-grained representations. Besides, we leverage two complementary alignments to further match cross-modal contents. Experimental results on two benchmarks reveal the challenges of VAR task and also demonstrate the advantages of our tailored method.

Abstract: In real-world traffic, there are various uncertainties and complexities in road and weather conditions. To solve the problem that the feature information of pole-like obstacles in complex environments is easily lost, resulting in low detection accuracy and low real-time performance, a multi-scale hybrid attention mechanism detection algorithm is proposed in this paper. First, the optimal transport function Monge-Kantorovich (MK) is incorporated not only to solve the problem of overlapping multiple prediction frames with optimal matching but also the MK function can be regularized to prevent model over-fitting; then, the features at different scales are up-sampled separately according to the optimized efficient multi-scale feature pyramid. Finally, the extraction of multi-scale feature space channel information is enhanced in complex environments based on the hybrid attention mechanism, which suppresses the irrelevant complex environment background information and focuses the feature information of pole-like obstacles. Meanwhile, this paper conducts real road test experiments in a variety of complex environments. The experimental results show that the detection precision, recall, and average precision of the method are 94.7%, 93.1%, and 97.4%, respectively, and the detection frame rate is 400 f/s. This research method can detect pole-like obstacles in a complex road environment in real time and accurately, which further promotes innovation and progress in the field of automatic driving.

Abstract: Dynamic vision sensors or event cameras provide rich complementary information for video frame interpolation. Existing state-of-the-art methods follow the paradigm of combining both synthesis-based and warping networks. However, few of those methods fully respect the intrinsic characteristics of events streams. Given that event cameras only encode intensity changes and polarity rather than color intensities, estimating optical flow from events is arguably more difficult than from RGB information. We therefore propose to incorporate RGB information in an event-guided optical flow refinement strategy. Moreover, in light of the quasi-continuous nature of the time signals provided by event cameras, we propose a divide-and-conquer strategy in which event-based intermediate frame synthesis happens incrementally in multiple simplified stages rather than in a single, long stage. Extensive experiments on both synthetic and real-world datasets show that these modifications lead to more reliable and realistic intermediate frame results than previous video frame interpolation methods. Our findings underline that a careful consideration of event characteristics such as high temporal density and elevated noise benefits interpolation accuracy.

Abstract: One little-explored frontier of image generation and editing is the task of interpolating between two input images, a feature missing from all currently deployed image generation pipelines. We argue that such a feature can expand the creative applications of such models, and propose a method for zero-shot interpolation using latent diffusion models. We apply interpolation in the latent space at a sequence of decreasing noise levels, then perform denoising conditioned on interpolated text embeddings derived from textual inversion and (optionally) subject poses. For greater consistency, or to specify additional criteria, we can generate several candidates and use CLIP to select the highest quality image. We obtain convincing interpolations across diverse subject poses, image styles, and image content, and show that standard quantitative metrics such as FID are insufficient to measure the quality of an interpolation. Code and data are available at https://clintonjwang.github.io/interpolation.

Abstract: Document dewarping from a distorted camera-captured image is of great value for OCR and document understanding. The document boundary plays an important role which is more evident than the inner region in document dewarping. Current learning-based methods mainly focus on complete boundary cases, leading to poor document correction performance of documents with incomplete boundaries. In contrast to these methods, this paper proposes MataDoc, the first method focusing on arbitrary boundary document dewarping with margin and text aware regularizations. Specifically, we design the margin regularization by explicitly considering background consistency to enhance boundary perception. Moreover, we introduce word position consistency to keep text lines straight in rectified document images. To produce a comprehensive evaluation of MataDoc, we propose a novel benchmark ArbDoc, mainly consisting of document images with arbitrary boundaries in four typical scenarios. Extensive experiments confirm the superiority of MataDoc with consideration for the incomplete boundary on ArbDoc and also demonstrate the effectiveness of the proposed method on DocUNet, DIR300, and WarpDoc datasets.

Abstract: In this paper, we strive to answer the question "how to collaboratively learn convolutional neural network (CNN)-based and vision transformer (ViT)-based models by selecting and exchanging the reliable knowledge between them for semantic segmentation?" Accordingly, we propose an online knowledge distillation (KD) framework that can simultaneously learn compact yet effective CNN-based and ViT-based models with two key technical breakthroughs to take full advantage of CNNs and ViT while compensating their limitations. Firstly, we propose heterogeneous feature distillation (HFD) to improve students' consistency in low-layer feature space by mimicking heterogeneous features between CNNs and ViT. Secondly, to facilitate the two students to learn reliable knowledge from each other, we propose bidirectional selective distillation (BSD) that can dynamically transfer selective knowledge. This is achieved by 1) region-wise BSD determining the directions of knowledge transferred between the corresponding regions in the feature space and 2) pixel-wise BSD discerning which of the prediction knowledge to be transferred in the logit space. Extensive experiments on three benchmark datasets demonstrate that our proposed framework outperforms the state-of-the-art online distillation methods by a large margin, and shows its efficacy in learning collaboratively between ViT-based and CNN-based models.

Abstract: Contrastive learning based vision-language joint pre-training has emerged as a successful representation learning strategy. In this paper, we present a prototype representation learning framework incorporating both global and local alignment between medical images and reports. In contrast to standard global multi-modality alignment methods, we employ a local alignment module for fine-grained representation. Furthermore, a cross-modality conditional reconstruction module is designed to interchange information across modalities in the training phase by reconstructing masked images and reports. For reconstructing long reports, a sentence-wise prototype memory bank is constructed, enabling the network to focus on low-level localized visual and high-level clinical linguistic features. Additionally, a non-auto-regressive generation paradigm is proposed for reconstructing non-sequential reports. Experimental results on five downstream tasks, including supervised classification, zero-shot classification, image-to-text retrieval, semantic segmentation, and object detection, show the proposed method outperforms other state-of-the-art methods across multiple datasets and under different dataset size settings. The code is available at https://github.com/QtacierP/PRIOR.

Abstract: Deep learning techniques for medical image analysis usually suffer from the domain shift between source and target data. Most existing works focus on unsupervised domain adaptation (UDA). However, in practical applications, privacy issues are much more severe. For example, the data of different hospitals have domain shifts due to equipment problems, and data of the two domains cannot be available simultaneously because of privacy. In this challenge defined as Source-Free UDA, the previous UDA medical methods are limited. Although a variety of medical source-free unsupervised domain adaption (MSFUDA) methods have been proposed, we found they fall into an over-fitting dilemma called "longer training, worse performance." Therefore, we propose the Stable Learning (SL) strategy to address the dilemma. SL is a scalable method and can be integrated with other research, which consists of Weight Consolidation and Entropy Increase. First, we apply Weight Consolidation to retain domain-invariant knowledge and then we design Entropy Increase to avoid over-learning. Comparative experiments prove the effectiveness of SL. We also have done extensive ablation experiments. Besides, We will release codes including a variety of MSFUDA methods.

Abstract: Recent advancements in large-scale Vision Transformers have made significant strides in improving pre-trained models for medical image segmentation. However, these methods face a notable challenge in acquiring a substantial amount of pre-training data, particularly within the medical field. To address this limitation, we present Masked Multi-view with Swin Transformers (SwinMM), a novel multi-view pipeline for enabling accurate and data-efficient self-supervised medical image analysis. Our strategy harnesses the potential of multi-view information by incorporating two principal components. In the pre-training phase, we deploy a masked multi-view encoder devised to concurrently train masked multi-view observations through a range of diverse proxy tasks. These tasks span image reconstruction, rotation, contrastive learning, and a novel task that employs a mutual learning paradigm. This new task capitalizes on the consistency between predictions from various perspectives, enabling the extraction of hidden multi-view information from 3D medical data. In the fine-tuning stage, a cross-view decoder is developed to aggregate the multi-view information through a cross-attention block. Compared with the previous state-of-the-art self-supervised learning method Swin UNETR, SwinMM demonstrates a notable advantage on several medical image segmentation tasks. It allows for a smooth integration of multi-view information, significantly boosting both the accuracy and data-efficiency of the model. Code and models are available at https://github.com/UCSC-VLAA/SwinMM/.

Abstract: DETR-like models have significantly boosted the performance of detectors and even outperformed classical convolutional models. However, all tokens are treated equally without discrimination brings a redundant computational burden in the traditional encoder structure. The recent sparsification strategies exploit a subset of informative tokens to reduce attention complexity maintaining performance through the sparse encoder. But these methods tend to rely on unreliable model statistics. Moreover, simply reducing the token population hinders the detection performance to a large extent, limiting the application of these sparse models. We propose Focus-DETR, which focuses attention on more informative tokens for a better trade-off between computation efficiency and model accuracy. Specifically, we reconstruct the encoder with dual attention, which includes a token scoring mechanism that considers both localization and category semantic information of the objects from multi-scale feature maps. We efficiently abandon the background queries and enhance the semantic interaction of the fine-grained object queries based on the scores. Compared with the state-of-the-art sparse DETR-like detectors under the same setting, our Focus-DETR gets comparable complexity while achieving 50.4AP (+2.2) on COCO. The code is available at https://github.com/huawei-noah/noah-research/tree/master/Focus-DETR and https://gitee.com/mindspore/models/tree/master/research/cv/Focus-DETR.

Abstract: The discrimination of instance embeddings plays a vital role in associating instances across time for online video instance segmentation (VIS). Instance embedding learning is directly supervised by the contrastive loss computed upon the contrastive items (CIs), which are sets of anchor/positive/negative embeddings. Recent online VIS methods leverage CIs sourced from one reference frame only, which we argue is insufficient for learning highly discriminative embeddings. Intuitively, a possible strategy to enhance CIs is replicating the inference phase during training. To this end, we propose a simple yet effective training strategy, called Consistent Training for Online VIS (CTVIS), which devotes to aligning the training and inference pipelines in terms of building CIs. Specifically, CTVIS constructs CIs by referring inference the momentum-averaged embedding and the memory bank storage mechanisms, and adding noise to the relevant embeddings. Such an extension allows a reliable comparison between embeddings of current instances and the stable representations of historical instances, thereby conferring an advantage in modeling VIS challenges such as occlusion, re-identification, and deformation. Empirically, CTVIS outstrips the SOTA VIS models by up to +5.0 points on three VIS benchmarks, including YTVIS19 (55.1% AP), YTVIS21 (50.1% AP) and OVIS (35.5% AP). Furthermore, we find that pseudo-videos transformed from images can train robust models surpassing fully-supervised ones.

Abstract: The Fourier transform, serving as an explicit decomposition method for visual signals, has been employed to explain the out-of-distribution generalization behaviors of Convolutional Neural Networks (CNNs). Previous research and empirical studies have indicated that the amplitude spectrum plays a decisive role in CNN recognition, but it is susceptible to disturbance caused by distribution shifts. On the other hand, the phase spectrum preserves highly-structured spatial information, which is crucial for visual representation learning. In this paper, we aim to clarify the relationships between Domain Generalization (DG) and the frequency components by introducing a Fourier-based structural causal model. Specifically, we interpret the phase spectrum as semi-causal factors and the amplitude spectrum as non-causal factors. Building upon these observations, we propose Phase Match (PhaMa) to address DG problems. Our method introduces perturbations on the amplitude spectrum and establishes spatial relationships to match the phase components. Through experiments on multiple benchmarks, we demonstrate that our proposed method achieves state-of-the-art performance in domain generalization and out-of-distribution robustness tasks.

Abstract: Medical image segmentation has made significant progress when a large amount of labeled data are available. However, annotating medical image segmentation datasets is expensive due to the requirement of professional skills. Additionally, classes are often unevenly distributed in medical images, which severely affects the classification performance on minority classes. To address these problems, this paper proposes Co-Distribution Alignment (Co-DA) for semi-supervised medical image segmentation. Specifically, Co-DA aligns marginal predictions on unlabeled data to marginal predictions on labeled data in a class-wise manner with two differently initialized models before using the pseudo-labels generated by one model to supervise the other. Besides, we design an over-expectation cross-entropy loss for filtering the unlabeled pixels to reduce noise in their pseudo-labels. Quantitative and qualitative experiments on three public datasets demonstrate that the proposed approach outperforms existing state-of-the-art semi-supervised medical image segmentation methods on both the 2D CaDIS dataset and the 3D LGE-MRI and ACDC datasets, achieving an mIoU of 0.8515 with only 24% labeled data on CaDIS, and a Dice score of 0.8824 and 0.8773 with only 20% data on LGE-MRI and ACDC, respectively.

Abstract: One of the main challenges in LiDAR-based 3D object detection is that the sensors often fail to capture the complete spatial information about the objects due to long distance and occlusion. Two-stage detectors with point cloud completion approaches tackle this problem by adding more points to the regions of interest (RoIs) with a pre-trained network. However, these methods generate dense point clouds of objects for all region proposals, assuming that objects always exist in the RoIs. This leads to the indiscriminate point generation for incorrect proposals as well. Motivated by this, we propose Point Generation R-CNN (PG-RCNN), a novel end-to-end detector that generates semantic surface points of foreground objects for accurate detection. Our method uses a jointly trained RoI point generation module to process the contextual information of RoIs and estimate the complete shape and displacement of foreground objects. For every generated point, PG-RCNN assigns a semantic feature that indicates the estimated foreground probability. Extensive experiments show that the point clouds generated by our method provide geometrically and semantically rich information for refining false positive and misaligned proposals. PG-RCNN achieves competitive performance on the KITTI benchmark, with significantly fewer parameters than state-of-the-art models. The code is available at https://github.com/quotation2520/PG-RCNN.

Abstract: Inspired by the fact that the matrix formulated by nonlocal similar patches in a natural image is of low rank, the rank approximation issue have been extensively investigated over the past decades, among which weighted nuclear norm minimization (WNNM) and weighted Schatten $p$-norm minimization (WSNM) are two prevailing methods have shown great superiority in various image restoration (IR) problems. Due to the physical characteristic of color images, color image restoration (CIR) is often a much more difficult task than its grayscale image counterpart. However, when applied to CIR, the traditional WNNM/WSNM method only processes three color channels individually and fails to consider their cross-channel correlations. Very recently, a quaternion-based WNNM approach (QWNNM) has been developed to mitigate this issue, which is capable of representing the color image as a whole in the quaternion domain and preserving the inherent correlation among the three color channels. Despite its empirical success, unfortunately, the convergence behavior of QWNNM has not been strictly studied yet. In this paper, on the one side, we extend the WSNM into quaternion domain and correspondingly propose a novel quaternion-based WSNM model (QWSNM) for tackling the CIR problems. Extensive experiments on two representative CIR tasks, including color image denoising and deblurring, demonstrate that the proposed QWSNM method performs favorably against many state-of-the-art alternatives, in both quantitative and qualitative evaluations. On the other side, more importantly, we preliminarily provide a theoretical convergence analysis, that is, by modifying the quaternion alternating direction method of multipliers (QADMM) through a simple continuation strategy, we theoretically prove that both the solution sequences generated by the QWNNM and QWSNM have fixed-point convergence guarantees.

Abstract: Deploying deep learning-based applications in specialized domains like the aircraft production industry typically suffers from the training data availability problem. Only a few datasets represent non-everyday objects, situations, and tasks. Recent advantages in research around Vision Foundation Models (VFM) opened a new area of tasks and models with high generalization capabilities in non-semantic and semantic predictions. As recently demonstrated by the Segment Anything Project, exploiting VFM's zero-shot capabilities is a promising direction in tackling the boundaries spanned by data, context, and sensor variety. Although, investigating its application within specific domains is subject to ongoing research. This paper contributes here by surveying applications of the SAM in aircraft production-specific use cases. We include manufacturing, intralogistics, as well as maintenance, repair, and overhaul processes, also representing a variety of other neighboring industrial domains. Besides presenting the various use cases, we further discuss the injection of domain knowledge.

Abstract: In past decade, previous balanced datasets have been used to advance algorithms for classification, object detection, semantic segmentation, and anomaly detection in industrial applications. Specifically, for condition-based maintenance, automating visual inspection is crucial to ensure high quality. Deterioration prognostic attempts to optimize the fine decision process for predictive maintenance and proactive repair. In civil infrastructure and living environment, damage data mining cannot avoid the imbalanced data issue because of rare unseen events and high quality status by improved operations. For visual inspection, deteriorated class acquired from the surface of concrete and steel components are occasionally imbalanced. From numerous related surveys, we summarize that imbalanced data problems can be categorized into four types; 1) missing range of target and label valuables, 2) majority-minority class imbalance, 3) foreground-background of spatial imbalance, 4) long-tailed class of pixel-wise imbalance. Since 2015, there has been many imbalanced studies using deep learning approaches that includes regression, image classification, object detection, semantic segmentation. However, anomaly detection for imbalanced data is not yet well known. In the study, we highlight one-class anomaly detection application whether anomalous class or not, and demonstrate clear examples on imbalanced vision datasets: wooden, concrete deterioration, and disaster damage. We provide key results on damage vision mining advantage, hypothesizing that the more effective range of positive ratio, the higher accuracy gain of anomaly detection application. Finally, the applicability of the damage learning methods, limitations, and future works are mentioned.

Abstract: Self-supervised learning of visual representations has been focusing on learning content features, which do not capture object motion or location, and focus on identifying and differentiating objects in images and videos. On the other hand, optical flow estimation is a task that does not involve understanding the content of the images on which it is estimated. We unify the two approaches and introduce MC-JEPA, a joint-embedding predictive architecture and self-supervised learning approach to jointly learn optical flow and content features within a shared encoder, demonstrating that the two associated objectives; the optical flow estimation objective and the self-supervised learning objective; benefit from each other and thus learn content features that incorporate motion information. The proposed approach achieves performance on-par with existing unsupervised optical flow benchmarks, as well as with common self-supervised learning approaches on downstream tasks such as semantic segmentation of images and videos.

Abstract: CT images corrupted by metal artifacts have serious negative effects on clinical diagnosis. Considering the difficulty of collecting paired data with ground truth in clinical settings, unsupervised methods for metal artifact reduction are of high interest. However, it is difficult for previous unsupervised methods to retain structural information from CT images while handling the non-local characteristics of metal artifacts. To address these challenges, we proposed a novel Dense Transformer based Enhanced Coding Network (DTEC-Net) for unsupervised metal artifact reduction. Specifically, we introduce a Hierarchical Disentangling Encoder, supported by the high-order dense process, and transformer to obtain densely encoded sequences with long-range correspondence. Then, we present a second-order disentanglement method to improve the dense sequence's decoding process. Extensive experiments and model discussions illustrate DTEC-Net's effectiveness, which outperforms the previous state-of-the-art methods on a benchmark dataset, and greatly reduces metal artifacts while restoring richer texture details.

Abstract: Neural Radiance Fields (NeRFs) have gained widespread recognition as a highly effective technique for representing 3D reconstructions of objects and scenes derived from sets of images. Despite their efficiency, NeRF models can pose challenges in certain scenarios such as vehicle inspection, where the lack of sufficient data or the presence of challenging elements (e.g. reflections) strongly impact the accuracy of the reconstruction. To this aim, we introduce CarPatch, a novel synthetic benchmark of vehicles. In addition to a set of images annotated with their intrinsic and extrinsic camera parameters, the corresponding depth maps and semantic segmentation masks have been generated for each view. Global and part-based metrics have been defined and used to evaluate, compare, and better characterize some state-of-the-art techniques. The dataset is publicly released at https://aimagelab.ing.unimore.it/go/carpatch and can be used as an evaluation guide and as a baseline for future work on this challenging topic.

Abstract: Unsupervised anomaly detection in medical images such as chest radiographs is stepping into the spotlight as it mitigates the scarcity of the labor-intensive and costly expert annotation of anomaly data. However, nearly all existing methods are formulated as a one-class classification trained only on representations from the normal class and discard a potentially significant portion of the unlabeled data. This paper focuses on a more practical setting, dual distribution anomaly detection for chest X-rays, using the entire training data, including both normal and unlabeled images. Inspired by a modern self-supervised vision transformer model trained using partial image inputs to reconstruct missing image regions -- we propose AMAE, a two-stage algorithm for adaptation of the pre-trained masked autoencoder (MAE). Starting from MAE initialization, AMAE first creates synthetic anomalies from only normal training images and trains a lightweight classifier on frozen transformer features. Subsequently, we propose an adaptation strategy to leverage unlabeled images containing anomalies. The adaptation scheme is accomplished by assigning pseudo-labels to unlabeled images and using two separate MAE based modules to model the normative and anomalous distributions of pseudo-labeled images. The effectiveness of the proposed adaptation strategy is evaluated with different anomaly ratios in an unlabeled training set. AMAE leads to consistent performance gains over competing self-supervised and dual distribution anomaly detection methods, setting the new state-of-the-art on three public chest X-ray benchmarks: RSNA, NIH-CXR, and VinDr-CXR.

Abstract: Humans are highly adaptable, swiftly switching between different modes to progressively handle different tasks, situations and contexts. In Human-object interaction (HOI) activities, these modes can be attributed to two mechanisms: (1) the large-scale consistent plan for the whole activity and (2) the small-scale children interactive actions that start and end along the timeline. While neuroscience and cognitive science have confirmed this multi-mechanism nature of human behavior, machine modeling approaches for human motion are trailing behind. While attempted to use gradually morphing structures (e.g., graph attention networks) to model the dynamic HOI patterns, they miss the expeditious and discrete mode-switching nature of the human motion. To bridge that gap, this work proposes to model two concurrent mechanisms that jointly control human motion: the Persistent process that runs continually on the global scale, and the Transient sub-processes that operate intermittently on the local context of the human while interacting with objects. These two mechanisms form an interactive Persistent-Transient Duality that synergistically governs the activity sequences. We model this conceptual duality by a parent-child neural network of Persistent and Transient channels with a dedicated neural module for dynamic mechanism switching. The framework is trialed on HOI motion forecasting. On two rich datasets and a wide variety of settings, the model consistently delivers superior performances, proving its suitability for the challenge.

Abstract: Practical object detection application can lose its effectiveness on image inputs with natural distribution shifts. This problem leads the research community to pay more attention on the robustness of detectors under Out-Of-Distribution (OOD) inputs. Existing works construct datasets to benchmark the detector's OOD robustness for a specific application scenario, e.g., Autonomous Driving. However, these datasets lack universality and are hard to benchmark general detectors built on common tasks such as COCO. To give a more comprehensive robustness assessment, we introduce COCO-O(ut-of-distribution), a test dataset based on COCO with 6 types of natural distribution shifts. COCO-O has a large distribution gap with training data and results in a significant 55.7% relative performance drop on a Faster R-CNN detector. We leverage COCO-O to conduct experiments on more than 100 modern object detectors to investigate if their improvements are credible or just over-fitting to the COCO test set. Unfortunately, most classic detectors in early years do not exhibit strong OOD generalization. We further study the robustness effect on recent breakthroughs of detector's architecture design, augmentation and pre-training techniques. Some empirical findings are revealed: 1) Compared with detection head or neck, backbone is the most important part for robustness; 2) An end-to-end detection transformer design brings no enhancement, and may even reduce robustness; 3) Large-scale foundation models have made a great leap on robust object detection. We hope our COCO-O could provide a rich testbed for robustness study of object detection. The dataset will be available at \url{https://github.com/alibaba/easyrobust/tree/main/benchmarks/coco_o}.

Abstract: CLIP has become a promising language-supervised visual pre-training framework and achieves excellent performance over a wide range of tasks. This paper aims to distill small CLIP models supervised by a large teacher CLIP model. We propose several distillation strategies, including relation, feature, gradient and contrastive paradigm, to examine the impact on CLIP distillation. We show that the simplest feature mimicry with MSE loss performs best. Moreover, interactive contrastive learning and relation-based distillation are also critical in performance improvement. We apply the unified method to distill several student networks trained on 15 million (image, text) pairs. Distillation improves the student CLIP models consistently over zero-shot ImageNet classification and cross-modal retrieval benchmarks. We hope our empirical study will become an important baseline for future CLIP distillation research. The code is available at \url{https://github.com/winycg/CLIP-KD}.

Abstract: Depth estimation is one of the essential tasks to be addressed when creating mobile autonomous systems. While monocular depth estimation methods have improved in recent times, depth completion provides more accurate and reliable depth maps by additionally using sparse depth information from other sensors such as LiDAR. However, current methods are specifically trained for a single LiDAR sensor. As the scanning pattern differs between sensors, every new sensor would require re-training a specialized depth completion model, which is computationally inefficient and not flexible. Therefore, we propose to dynamically adapt the depth completion model to the used sensor type enabling LiDAR adaptive depth completion. Specifically, we propose a meta depth completion network that uses data patterns derived from the data to learn a task network to alter weights of the main depth completion network to solve a given depth completion task effectively. The method demonstrates a strong capability to work on multiple LiDAR scanning patterns and can also generalize to scanning patterns that are unseen during training. While using a single model, our method yields significantly better results than a non-adaptive baseline trained on different LiDAR patterns. It outperforms LiDAR-specific expert models for very sparse cases. These advantages allow flexible deployment of a single depth completion model on different sensors, which could also prove valuable to process the input of nascent LiDAR technology with adaptive instead of fixed scanning patterns.

Abstract: Video stabilization refers to the problem of transforming a shaky video into a visually pleasing one. The question of how to strike a good trade-off between visual quality and computational speed has remained one of the open challenges in video stabilization. Inspired by the analogy between wobbly frames and jigsaw puzzles, we propose an iterative optimization-based learning approach using synthetic datasets for video stabilization, which consists of two interacting submodules: motion trajectory smoothing and full-frame outpainting. First, we develop a two-level (coarse-to-fine) stabilizing algorithm based on the probabilistic flow field. The confidence map associated with the estimated optical flow is exploited to guide the search for shared regions through backpropagation. Second, we take a divide-and-conquer approach and propose a novel multiframe fusion strategy to render full-frame stabilized views. An important new insight brought about by our iterative optimization approach is that the target video can be interpreted as the fixed point of nonlinear mapping for video stabilization. We formulate video stabilization as a problem of minimizing the amount of jerkiness in motion trajectories, which guarantees convergence with the help of fixed-point theory. Extensive experimental results are reported to demonstrate the superiority of the proposed approach in terms of computational speed and visual quality. The code will be available on GitHub.

Abstract: Medical imaging is a key component in clinical diagnosis, treatment planning and clinical trial design, accounting for almost 90% of all healthcare data. CNNs achieved performance gains in medical image analysis (MIA) over the last years. CNNs can efficiently model local pixel interactions and be trained on small-scale MI data. The main disadvantage of typical CNN models is that they ignore global pixel relationships within images, which limits their generalisation ability to understand out-of-distribution data with different 'global' information. The recent progress of Artificial Intelligence gave rise to Transformers, which can learn global relationships from data. However, full Transformer models need to be trained on large-scale data and involve tremendous computational complexity. Attention and Transformer compartments (Transf/Attention) which can well maintain properties for modelling global relationships, have been proposed as lighter alternatives of full Transformers. Recently, there is an increasing trend to co-pollinate complementary local-global properties from CNN and Transf/Attention architectures, which led to a new era of hybrid models. The past years have witnessed substantial growth in hybrid CNN-Transf/Attention models across diverse MIA problems. In this systematic review, we survey existing hybrid CNN-Transf/Attention models, review and unravel key architectural designs, analyse breakthroughs, and evaluate current and future opportunities as well as challenges. We also introduced a comprehensive analysis framework on generalisation opportunities of scientific and clinical impact, based on which new data-driven domain generalisation and adaptation methods can be stimulated.

Abstract: Detecting objects based on language descriptions is a popular task that includes Open-Vocabulary object Detection (OVD) and Referring Expression Comprehension (REC). In this paper, we advance them to a more practical setting called Described Object Detection (DOD) by expanding category names to flexible language expressions for OVD and overcoming the limitation of REC to only grounding the pre-existing object. We establish the research foundation for DOD tasks by constructing a Description Detection Dataset ($D^3$), featuring flexible language expressions and annotating all described objects without omission. By evaluating previous SOTA methods on $D^3$, we find some troublemakers that fail current REC, OVD, and bi-functional methods. REC methods struggle with confidence scores, rejecting negative instances, and multi-target scenarios, while OVD methods face constraints with long and complex descriptions. Recent bi-functional methods also do not work well on DOD due to their separated training procedures and inference strategies for REC and OVD tasks. Building upon the aforementioned findings, we propose a baseline that largely improves REC methods by reconstructing the training data and introducing a binary classification sub-task, outperforming existing methods. Data and code is available at https://github.com/shikras/d-cube.

Abstract: This report presents the technical details of our approach for the EPIC-Kitchens-100 Unsupervised Domain Adaptation (UDA) Challenge in Action Recognition. Our approach is based on the idea that the order in which actions are performed is similar between the source and target domains. Based on this, we generate a modified sequence by randomly combining actions from the source and target domains. As only unlabelled target data are available under the UDA setting, we use a standard pseudo-labeling strategy for extracting action labels for the target. We then ask the network to predict the resulting action sequence. This allows to integrate information from both domains during training and to achieve better transfer results on target. Additionally, to better incorporate sequence information, we use a language model to filter unlikely sequences. Lastly, we employed a co-occurrence matrix to eliminate unseen combinations of verbs and nouns. Our submission, labeled as 'sshayan', can be found on the leaderboard, where it currently holds the 2nd position for 'verb' and the 4th position for both 'noun' and 'action'.

Abstract: Long-term activity forecasting is an especially challenging research problem because it requires understanding the temporal relationships between observed actions, as well as the variability and complexity of human activities. Despite relying on strong supervision via expensive human annotations, state-of-the-art forecasting approaches often generalize poorly to unseen data. To alleviate this issue, we propose Multiscale Video Pretraining (MVP), a novel self-supervised pretraining approach that learns robust representations for forecasting by learning to predict contextualized representations of future video clips over multiple timescales. MVP is based on our observation that actions in videos have a multiscale nature, where atomic actions typically occur at a short timescale and more complex actions may span longer timescales. We compare MVP to state-of-the-art self-supervised video learning approaches on downstream long-term forecasting tasks including long-term action anticipation and video summary prediction. Our comprehensive experiments across the Ego4D and Epic-Kitchens-55/100 datasets demonstrate that MVP out-performs state-of-the-art methods by significant margins. Notably, MVP obtains a relative performance gain of over 20% accuracy in video summary forecasting over existing methods.

Abstract: Intracerebral hemorrhage (ICH) is the second most common and deadliest form of stroke. Despite medical advances, predicting treat ment outcomes for ICH remains a challenge. This paper proposes a novel prognostic model that utilizes both imaging and tabular data to predict treatment outcome for ICH. Our model is trained on observational data collected from non-randomized controlled trials, providing reliable predictions of treatment success. Specifically, we propose to employ a variational autoencoder model to generate a low-dimensional prognostic score, which can effectively address the selection bias resulting from the non-randomized controlled trials. Importantly, we develop a variational distributions combination module that combines the information from imaging data, non-imaging clinical data, and treatment assignment to accurately generate the prognostic score. We conducted extensive experiments on a real-world clinical dataset of intracerebral hemorrhage. Our proposed method demonstrates a substantial improvement in treatment outcome prediction compared to existing state-of-the-art approaches. Code is available at https://github.com/med-air/TOP-GPM

Abstract: Despite the success of diffusion models (DMs), we still lack a thorough understanding of their latent space. To understand the latent space $\mathbf{x}_t \in \mathcal{X}$, we analyze them from a geometrical perspective. Specifically, we utilize the pullback metric to find the local latent basis in $\mathcal{X}$ and their corresponding local tangent basis in $\mathcal{H}$, the intermediate feature maps of DMs. The discovered latent basis enables unsupervised image editing capability through latent space traversal. We investigate the discovered structure from two perspectives. First, we examine how geometric structure evolves over diffusion timesteps. Through analysis, we show that 1) the model focuses on low-frequency components early in the generative process and attunes to high-frequency details later; 2) At early timesteps, different samples share similar tangent spaces; and 3) The simpler datasets that DMs trained on, the more consistent the tangent space for each timestep. Second, we investigate how the geometric structure changes based on text conditioning in Stable Diffusion. The results show that 1) similar prompts yield comparable tangent spaces; and 2) the model depends less on text conditions in later timesteps. To the best of our knowledge, this paper is the first to present image editing through $\mathbf{x}$-space traversal and provide thorough analyses of the latent structure of DMs.

Abstract: Since the training data for the target model in a data-free black-box attack is not available, most recent schemes utilize GANs to generate data for training substitute model. However, these GANs-based schemes suffer from low training efficiency as the generator needs to be retrained for each target model during the substitute training process, as well as low generation quality. To overcome these limitations, we consider utilizing the diffusion model to generate data, and propose a data-free black-box attack scheme based on diffusion model to improve the efficiency and accuracy of substitute training. Despite the data generated by the diffusion model exhibits high quality, it presents diverse domain distributions and contains many samples that do not meet the discriminative criteria of the target model. To further facilitate the diffusion model to generate data suitable for the target model, we propose a Latent Code Augmentation (LCA) method to guide the diffusion model in generating data. With the guidance of LCA, the data generated by the diffusion model not only meets the discriminative criteria of the target model but also exhibits high diversity. By utilizing this data, it is possible to train substitute model that closely resemble the target model more efficiently. Extensive experiments demonstrate that our LCA achieves higher attack success rates and requires fewer query budgets compared to GANs-based schemes for different target models.

Abstract: Event-based sensors, with their high temporal resolution (1us) and dynamical range (120dB), have the potential to be deployed in high-speed platforms such as vehicles and drones. However, the highly sparse and fluctuating nature of events poses challenges for conventional object detection techniques based on Artificial Neural Networks (ANNs). In contrast, Spiking Neural Networks (SNNs) are well-suited for representing event-based data due to their inherent temporal dynamics. In particular, we demonstrate that the membrane potential dynamics can modulate network activity upon fluctuating events and strengthen features of sparse input. In addition, the spike-triggered adaptive threshold can stabilize training which further improves network performance. Based on this, we develop an efficient spiking feature pyramid network for event-based object detection. Our proposed SNN outperforms previous SNNs and sophisticated ANNs with attention mechanisms, achieving a mean average precision (map50) of 47.7% on the Gen1 benchmark dataset. This result significantly surpasses the previous best SNN by 9.7% and demonstrates the potential of SNNs for event-based vision. Our model has a concise architecture while maintaining high accuracy and much lower computation cost as a result of sparse computation. Our code will be publicly available.

Abstract: Vision-and-language navigation (VLN) enables the agent to navigate to a remote location following the natural language instruction in 3D environments. To represent the previously visited environment, most approaches for VLN implement memory using recurrent states, topological maps, or top-down semantic maps. In contrast to these approaches, we build the top-down egocentric and dynamically growing Grid Memory Map (i.e., GridMM) to structure the visited environment. From a global perspective, historical observations are projected into a unified grid map in a top-down view, which can better represent the spatial relations of the environment. From a local perspective, we further propose an instruction relevance aggregation method to capture fine-grained visual clues in each grid region. Extensive experiments are conducted on both the REVERIE, R2R, SOON datasets in the discrete environments, and the R2R-CE dataset in the continuous environments, showing the superiority of our proposed method.

Abstract: Recently, the editing of neural radiance fields (NeRFs) has gained considerable attention, but most prior works focus on static scenes while research on the appearance editing of dynamic scenes is relatively lacking. In this paper, we propose a novel framework to edit the local appearance of dynamic NeRFs by manipulating pixels in a single frame of training video. Specifically, to locally edit the appearance of dynamic NeRFs while preserving unedited regions, we introduce a local surface representation of the edited region, which can be inserted into and rendered along with the original NeRF and warped to arbitrary other frames through a learned invertible motion representation network. By employing our method, users without professional expertise can easily add desired content to the appearance of a dynamic scene. We extensively evaluate our approach on various scenes and show that our approach achieves spatially and temporally consistent editing results. Notably, our approach is versatile and applicable to different variants of dynamic NeRF representations.

Abstract: The accelerated adoption of digital pathology and advances in deep learning have enabled the development of powerful models for various pathology tasks across a diverse array of diseases and patient cohorts. However, model training is often difficult due to label scarcity in the medical domain and the model's usage is limited by the specific task and disease for which it is trained. Additionally, most models in histopathology leverage only image data, a stark contrast to how humans teach each other and reason about histopathologic entities. We introduce CONtrastive learning from Captions for Histopathology (CONCH), a visual-language foundation model developed using diverse sources of histopathology images, biomedical text, and notably over 1.17 million image-caption pairs via task-agnostic pretraining. Evaluated on a suite of 13 diverse benchmarks, CONCH can be transferred to a wide range of downstream tasks involving either or both histopathology images and text, achieving state-of-the-art performance on histology image classification, segmentation, captioning, text-to-image and image-to-text retrieval. CONCH represents a substantial leap over concurrent visual-language pretrained systems for histopathology, with the potential to directly facilitate a wide array of machine learning-based workflows requiring minimal or no further supervised fine-tuning.

Abstract: With rapid advancements in depth sensors and deep learning, skeleton-based person re-identification (re-ID) models have recently achieved remarkable progress with many advantages. Most existing solutions learn single-level skeleton features from body joints with the assumption of equal skeleton importance, while they typically lack the ability to exploit more informative skeleton features from various levels such as limb level with more global body patterns. The label dependency of these methods also limits their flexibility in learning more general skeleton representations. This paper proposes a generic unsupervised Hierarchical skeleton Meta-Prototype Contrastive learning (Hi-MPC) approach with Hard Skeleton Mining (HSM) for person re-ID with unlabeled 3D skeletons. Firstly, we construct hierarchical representations of skeletons to model coarse-to-fine body and motion features from the levels of body joints, components, and limbs. Then a hierarchical meta-prototype contrastive learning model is proposed to cluster and contrast the most typical skeleton features ("prototypes") from different-level skeletons. By converting original prototypes into meta-prototypes with multiple homogeneous transformations, we induce the model to learn the inherent consistency of prototypes to capture more effective skeleton features for person re-ID. Furthermore, we devise a hard skeleton mining mechanism to adaptively infer the informative importance of each skeleton, so as to focus on harder skeletons to learn more discriminative skeleton representations. Extensive evaluations on five datasets demonstrate that our approach outperforms a wide variety of state-of-the-art skeleton-based methods. We further show the general applicability of our method to cross-view person re-ID and RGB-based scenarios with estimated skeletons.

Abstract: Text-to-video retrieval systems have recently made significant progress by utilizing pre-trained models trained on large-scale image-text pairs. However, most of the latest methods primarily focus on the video modality while disregarding the audio signal for this task. Nevertheless, a recent advancement by ECLIPSE has improved long-range text-to-video retrieval by developing an audiovisual video representation. Nonetheless, the objective of the text-to-video retrieval task is to capture the complementary audio and video information that is pertinent to the text query rather than simply achieving better audio and video alignment. To address this issue, we introduce TEFAL, a TExt-conditioned Feature ALignment method that produces both audio and video representations conditioned on the text query. Instead of using only an audiovisual attention block, which could suppress the audio information relevant to the text query, our approach employs two independent cross-modal attention blocks that enable the text to attend to the audio and video representations separately. Our proposed method's efficacy is demonstrated on four benchmark datasets that include audio: MSR-VTT, LSMDC, VATEX, and Charades, and achieves better than state-of-the-art performance consistently across the four datasets. This is attributed to the additional text-query-conditioned audio representation and the complementary information it adds to the text-query-conditioned video representation.

Abstract: Humans effortlessly grasp the connection between sketches and real-world objects, even when these sketches are far from realistic. Moreover, human sketch understanding goes beyond categorization -- critically, it also entails understanding how individual elements within a sketch correspond to parts of the physical world it represents. What are the computational ingredients needed to support this ability? Towards answering this question, we make two contributions: first, we introduce a new sketch-photo correspondence benchmark, $\textit{PSC6k}$, containing 150K annotations of 6250 sketch-photo pairs across 125 object categories, augmenting the existing Sketchy dataset with fine-grained correspondence metadata. Second, we propose a self-supervised method for learning dense correspondences between sketch-photo pairs, building upon recent advances in correspondence learning for pairs of photos. Our model uses a spatial transformer network to estimate the warp flow between latent representations of a sketch and photo extracted by a contrastive learning-based ConvNet backbone. We found that this approach outperformed several strong baselines and produced predictions that were quantitatively consistent with other warp-based methods. However, our benchmark also revealed systematic differences between predictions of the suite of models we tested and those of humans. Taken together, our work suggests a promising path towards developing artificial systems that achieve more human-like understanding of visual images at different levels of abstraction. Project page: https://photo-sketch-correspondence.github.io

Abstract: Volcanic eruptions emit ash that can be harmful to human health and cause damage to infrastructure, economic activities and the environment. The delimitation of ash clouds allows to know their behavior and dispersion, which helps in the prevention and mitigation of this phenomenon. Traditional methods take advantage of specialized software programs to process the bands or channels that compose the satellite images. However, their use is limited to experts and demands a lot of time and significant computational resources. In recent years, Artificial Intelligence has been a milestone in the computational treatment of complex problems in different areas. In particular, Deep Learning techniques allow automatic, fast and accurate processing of digital images. The present work proposes the use of the Pix2Pix model, a type of generative adversarial network that, once trained, learns the mapping of input images to output images. The architecture of such a network consisting of a generator and a discriminator provides the versatility needed to produce black and white ash cloud images from multispectral satellite images. The evaluation of the model, based on loss and accuracy plots, a confusion matrix, and visual inspection, indicates a satisfactory solution for accurate ash cloud delineation, applicable in any area of the world and becomes a useful tool in risk management.

Abstract: In this paper, we propose a new operator, called 3D DeFormable Attention (DFA3D), for 2D-to-3D feature lifting, which transforms multi-view 2D image features into a unified 3D space for 3D object detection. Existing feature lifting approaches, such as Lift-Splat-based and 2D attention-based, either use estimated depth to get pseudo LiDAR features and then splat them to a 3D space, which is a one-pass operation without feature refinement, or ignore depth and lift features by 2D attention mechanisms, which achieve finer semantics while suffering from a depth ambiguity problem. In contrast, our DFA3D-based method first leverages the estimated depth to expand each view's 2D feature map to 3D and then utilizes DFA3D to aggregate features from the expanded 3D feature maps. With the help of DFA3D, the depth ambiguity problem can be effectively alleviated from the root, and the lifted features can be progressively refined layer by layer, thanks to the Transformer-like architecture. In addition, we propose a mathematically equivalent implementation of DFA3D which can significantly improve its memory efficiency and computational speed. We integrate DFA3D into several methods that use 2D attention-based feature lifting with only a few modifications in code and evaluate on the nuScenes dataset. The experiment results show a consistent improvement of +1.41\% mAP on average, and up to +15.1\% mAP improvement when high-quality depth information is available, demonstrating the superiority, applicability, and huge potential of DFA3D. The code is available at https://github.com/IDEA-Research/3D-deformable-attention.git.

Abstract: Prompt engineering is a technique that involves augmenting a large pre-trained model with task-specific hints, known as prompts, to adapt the model to new tasks. Prompts can be created manually as natural language instructions or generated automatically as either natural language instructions or vector representations. Prompt engineering enables the ability to perform predictions based solely on prompts without updating model parameters, and the easier application of large pre-trained models in real-world tasks. In past years, Prompt engineering has been well-studied in natural language processing. Recently, it has also been intensively studied in vision-language modeling. However, there is currently a lack of a systematic overview of prompt engineering on pre-trained vision-language models. This paper aims to provide a comprehensive survey of cutting-edge research in prompt engineering on three types of vision-language models: multimodal-to-text generation models (e.g. Flamingo), image-text matching models (e.g. CLIP), and text-to-image generation models (e.g. Stable Diffusion). For each type of model, a brief model summary, prompting methods, prompting-based applications, and the corresponding responsibility and integrity issues are summarized and discussed. Furthermore, the commonalities and differences between prompting on vision-language models, language models, and vision models are also discussed. The challenges, future directions, and research opportunities are summarized to foster future research on this topic.

Abstract: Large language models (LLMs) and Vision-Language Models (VLMs) have been proven to excel at multiple tasks, such as commonsense reasoning. Powerful as these models can be, they are not grounded in the 3D physical world, which involves richer concepts such as spatial relationships, affordances, physics, layout, and so on. In this work, we propose to inject the 3D world into large language models and introduce a whole new family of 3D-LLMs. Specifically, 3D-LLMs can take 3D point clouds and their features as input and perform a diverse set of 3D-related tasks, including captioning, dense captioning, 3D question answering, task decomposition, 3D grounding, 3D-assisted dialog, navigation, and so on. Using three types of prompting mechanisms that we design, we are able to collect over 300k 3D-language data covering these tasks. To efficiently train 3D-LLMs, we first utilize a 3D feature extractor that obtains 3D features from rendered multi- view images. Then, we use 2D VLMs as our backbones to train our 3D-LLMs. By introducing a 3D localization mechanism, 3D-LLMs can better capture 3D spatial information. Experiments on ScanQA show that our model outperforms state-of-the-art baselines by a large margin (e.g., the BLEU-1 score surpasses state-of-the-art score by 9%). Furthermore, experiments on our held-in datasets for 3D captioning, task composition, and 3D-assisted dialogue show that our model outperforms 2D VLMs. Qualitative examples also show that our model could perform more tasks beyond the scope of existing LLMs and VLMs. Project Page: : https://vis-www.cs.umass.edu/3dllm/.