Computer Vision and Pattern Recognition (cs.CV)

Abstract: The process of annotating histological gigapixel-sized whole slide images (WSIs) at the pixel level for the purpose of training a supervised segmentation model is time-consuming. Region-based active learning (AL) involves training the model on a limited number of annotated image regions instead of requesting annotations of the entire images. These annotation regions are iteratively selected, with the goal of optimizing model performance while minimizing the annotated area. The standard method for region selection evaluates the informativeness of all square regions of a specified size and then selects a specific quantity of the most informative regions. We find that the efficiency of this method highly depends on the choice of AL step size (i.e., the combination of region size and the number of selected regions per WSI), and a suboptimal AL step size can result in redundant annotation requests or inflated computation costs. This paper introduces a novel technique for selecting annotation regions adaptively, mitigating the reliance on this AL hyperparameter. Specifically, we dynamically determine each region by first identifying an informative area and then detecting its optimal bounding box, as opposed to selecting regions of a uniform predefined shape and size as in the standard method. We evaluate our method using the task of breast cancer metastases segmentation on the public CAMELYON16 dataset and show that it consistently achieves higher sampling efficiency than the standard method across various AL step sizes. With only 2.6\% of tissue area annotated, we achieve full annotation performance and thereby substantially reduce the costs of annotating a WSI dataset. The source code is available at https://github.com/DeepMicroscopy/AdaptiveRegionSelection.

Abstract: The prediction of mild cognitive impairment (MCI) conversion to Alzheimer's disease (AD) is important for early treatment to prevent or slow the progression of AD. To accurately predict the MCI conversion to stable MCI or progressive MCI, we propose Triformer, a novel transformer-based framework with three specialized transformers to incorporate multi-model data. Triformer uses I) an image transformer to extract multi-view image features from medical scans, II) a clinical transformer to embed and correlate multi-modal clinical data, and III) a modality fusion transformer that produces an accurate prediction based on fusing the outputs from the image and clinical transformers. Triformer is evaluated on the Alzheimer's Disease Neuroimaging Initiative (ANDI)1 and ADNI2 datasets and outperforms previous state-of-the-art single and multi-modal methods.

Abstract: Domain generalization aims to learn a generalization model that can perform well on unseen test domains by only training on limited source domains. However, existing domain generalization approaches often bring in prediction-irrelevant noise or require the collection of domain labels. To address these challenges, we consider the domain generalization problem from a different perspective by categorizing underlying feature groups into domain-shared and domain-specific features. Nevertheless, the domain-specific features are difficult to be identified and distinguished from the input data. In this work, we propose DomaIn-SPEcific Liberating (DISPEL), a post-processing fine-grained masking approach that can filter out undefined and indistinguishable domain-specific features in the embedding space. Specifically, DISPEL utilizes a mask generator that produces a unique mask for each input data to filter domain-specific features. The DISPEL framework is highly flexible to be applied to any fine-tuned models. We derive a generalization error bound to guarantee the generalization performance by optimizing a designed objective loss. The experimental results on five benchmarks demonstrate DISPEL outperforms existing methods and can further generalize various algorithms.

Abstract: Most existing methods for text-based person retrieval focus on text-to-image person retrieval. Nevertheless, due to the lack of dynamic information provided by isolated frames, the performance is hampered when the person is obscured in isolated frames or variable motion details are given in the textual description. In this paper, we propose a new task called Text-to-Video Person Retrieval(TVPR) which aims to effectively overcome the limitations of isolated frames. Since there is no dataset or benchmark that describes person videos with natural language, we construct a large-scale cross-modal person video dataset containing detailed natural language annotations, such as person's appearance, actions and interactions with environment, etc., termed as Text-to-Video Person Re-identification (TVPReid) dataset, which will be publicly available. To this end, a Text-to-Video Person Retrieval Network (TVPRN) is proposed. Specifically, TVPRN acquires video representations by fusing visual and motion representations of person videos, which can deal with temporal occlusion and the absence of variable motion details in isolated frames. Meanwhile, we employ the pre-trained BERT to obtain caption representations and the relationship between caption and video representations to reveal the most relevant person videos. To evaluate the effectiveness of the proposed TVPRN, extensive experiments have been conducted on TVPReid dataset. To the best of our knowledge, TVPRN is the first successful attempt to use video for text-based person retrieval task and has achieved state-of-the-art performance on TVPReid dataset. The TVPReid dataset will be publicly available to benefit future research.

Abstract: Occlusion perturbation presents a significant challenge in person re-identification (re-ID), and existing methods that rely on external visual cues require additional computational resources and only consider the issue of missing information caused by occlusion. In this paper, we propose a simple yet effective framework, termed Erasing, Transforming, and Noising Defense Network (ETNDNet), which treats occlusion as a noise disturbance and solves occluded person re-ID from the perspective of adversarial defense. In the proposed ETNDNet, we introduce three strategies: Firstly, we randomly erase the feature map to create an adversarial representation with incomplete information, enabling adversarial learning of identity loss to protect the re-ID system from the disturbance of missing information. Secondly, we introduce random transformations to simulate the position misalignment caused by occlusion, training the extractor and classifier adversarially to learn robust representations immune to misaligned information. Thirdly, we perturb the feature map with random values to address noisy information introduced by obstacles and non-target pedestrians, and employ adversarial gaming in the re-ID system to enhance its resistance to occlusion noise. Without bells and whistles, ETNDNet has three key highlights: (i) it does not require any external modules with parameters, (ii) it effectively handles various issues caused by occlusion from obstacles and non-target pedestrians, and (iii) it designs the first GAN-based adversarial defense paradigm for occluded person re-ID. Extensive experiments on five public datasets fully demonstrate the effectiveness, superiority, and practicality of the proposed ETNDNet. The code will be released at \url{https://github.com/nengdong96/ETNDNet}.

Abstract: For smart vehicles driving through signalised intersections, it is crucial to determine whether the vehicle has right of way given the state of the traffic lights. To address this issue, camera based sensors can be used to determine whether the vehicle has permission to proceed straight, turn left or turn right. This paper proposes a novel end to end intersection right of way recognition model called LightFormer to generate right of way status for available driving directions in complex urban intersections. The model includes a spatial temporal inner structure with an attention mechanism, which incorporates features from past image to contribute to the classification of the current frame right of way status. In addition, a modified, multi weight arcface loss is introduced to enhance the model classification performance. Finally, the proposed LightFormer is trained and tested on two public traffic light datasets with manually augmented labels to demonstrate its effectiveness.

Abstract: Anomalies are rare and anomaly detection is often therefore framed as One-Class Classification (OCC), i.e. trained solely on normalcy. Leading OCC techniques constrain the latent representations of normal motions to limited volumes and detect as abnormal anything outside, which accounts satisfactorily for the openset'ness of anomalies. But normalcy shares the same openset'ness property, since humans can perform the same action in several ways, which the leading techniques neglect. We propose a novel generative model for video anomaly detection (VAD), which assumes that both normality and abnormality are multimodal. We consider skeletal representations and leverage state-of-the-art diffusion probabilistic models to generate multimodal future human poses. We contribute a novel conditioning on the past motion of people, and exploit the improved mode coverage capabilities of diffusion processes to generate different-but-plausible future motions. Upon the statistical aggregation of future modes, anomaly is detected when the generated set of motions is not pertinent to the actual future. We validate our model on 4 established benchmarks: UBnormal, HR-UBnormal, HR-STC, and HR-Avenue, with extensive experiments surpassing state-of-the-art results.

Abstract: Open-set image recognition is a challenging topic in computer vision. Most of the existing works in literature focus on learning more discriminative features from the input images, however, they are usually insensitive to the high- or low-frequency components in features, resulting in a decreasing performance on fine-grained image recognition. To address this problem, we propose a Complementary Frequency-varying Awareness Network that could better capture both high-frequency and low-frequency information, called CFAN. The proposed CFAN consists of three sequential modules: (i) a feature extraction module is introduced for learning preliminary features from the input images; (ii) a frequency-varying filtering module is designed to separate out both high- and low-frequency components from the preliminary features in the frequency domain via a frequency-adjustable filter; (iii) a complementary temporal aggregation module is designed for aggregating the high- and low-frequency components via two Long Short-Term Memory networks into discriminative features. Based on CFAN, we further propose an open-set fine-grained image recognition method, called CFAN-OSFGR, which learns image features via CFAN and classifies them via a linear classifier. Experimental results on 3 fine-grained datasets and 2 coarse-grained datasets demonstrate that CFAN-OSFGR performs significantly better than 9 state-of-the-art methods in most cases.

Abstract: While clinical trials are the state-of-the-art methods to assess the effect of new medication in a comparative manner, benchmarking in the field of medical image analysis is performed by so-called challenges. Recently, comprehensive analysis of multiple biomedical image analysis challenges revealed large discrepancies between the impact of challenges and quality control of the design and reporting standard. This work aims to follow up on these results and attempts to address the specific question of the reproducibility of the participants methods. In an effort to determine whether alternative interpretations of the method description may change the challenge ranking, we reproduced the algorithms submitted to the 2019 Robust Medical Image Segmentation Challenge (ROBUST-MIS). The leaderboard differed substantially between the original challenge and reimplementation, indicating that challenge rankings may not be sufficiently reproducible.

Abstract: While transformer models have been demonstrated to be effective for natural language processing tasks and high-level vision tasks, only a few attempts have been made to use powerful transformer models for single image super-resolution. Because transformer models have powerful representation capacity and the in-built self-attention mechanisms in transformer models help to leverage self-similarity prior in input low-resolution image to improve performance for single image super-resolution, we present a single image super-resolution model based on recent hybrid vision transformer of MaxViT, named as MaxSR. MaxSR consists of four parts, a shallow feature extraction block, multiple cascaded adaptive MaxViT blocks to extract deep hierarchical features and model global self-similarity from low-level features efficiently, a hierarchical feature fusion block, and finally a reconstruction block. The key component of MaxSR, i.e., adaptive MaxViT block, is based on MaxViT block which mixes MBConv with squeeze-and-excitation, block attention and grid attention. In order to achieve better global modelling of self-similarity in input low-resolution image, we improve block attention and grid attention in MaxViT block to adaptive block attention and adaptive grid attention which do self-attention inside each window across all grids and each grid across all windows respectively in the most efficient way. We instantiate proposed model for classical single image super-resolution (MaxSR) and lightweight single image super-resolution (MaxSR-light). Experiments show that our MaxSR and MaxSR-light establish new state-of-the-art performance efficiently.

Abstract: Curvilinear object segmentation is critical for many applications. However, manually annotating curvilinear objects is very time-consuming and error-prone, yielding insufficiently available annotated datasets for existing supervised methods and domain adaptation methods. This paper proposes a self-supervised curvilinear object segmentation method that learns robust and distinctive features from fractals and unlabeled images (FreeCOS). The key contributions include a novel Fractal-FDA synthesis (FFS) module and a geometric information alignment (GIA) approach. FFS generates curvilinear structures based on the parametric Fractal L-system and integrates the generated structures into unlabeled images to obtain synthetic training images via Fourier Domain Adaptation. GIA reduces the intensity differences between the synthetic and unlabeled images by comparing the intensity order of a given pixel to the values of its nearby neighbors. Such image alignment can explicitly remove the dependency on absolute intensity values and enhance the inherent geometric characteristics which are common in both synthetic and real images. In addition, GIA aligns features of synthetic and real images via the prediction space adaptation loss (PSAL) and the curvilinear mask contrastive loss (CMCL). Extensive experimental results on four public datasets, i.e., XCAD, DRIVE, STARE and CrackTree demonstrate that our method outperforms the state-of-the-art unsupervised methods, self-supervised methods and traditional methods by a large margin. The source code of this work is available at https://github.com/TY-Shi/FreeCOS.

Abstract: The foundation models based on pre-training technology have significantly advanced artificial intelligence from theoretical to practical applications. These models have facilitated the feasibility of computer-aided diagnosis for widespread use. Medical contrastive vision-language pre-training, which does not require human annotations, is an effective approach for guiding representation learning using description information in diagnostic reports. However, the effectiveness of pre-training is limited by the large-scale semantic overlap and shifting problems in medical field. To address these issues, we propose the Knowledge-Boosting Contrastive Vision-Language Pre-training framework (KoBo), which integrates clinical knowledge into the learning of vision-language semantic consistency. The framework uses an unbiased, open-set sample-wise knowledge representation to measure negative sample noise and supplement the correspondence between vision-language mutual information and clinical knowledge. Extensive experiments validate the effect of our framework on eight tasks including classification, segmentation, retrieval, and semantic relatedness, achieving comparable or better performance with the zero-shot or few-shot settings. Our code is open on https://github.com/ChenXiaoFei-CS/KoBo.

Abstract: Support vector machines (SVMs) are a well-established classifier effectively deployed in an array of pattern recognition and classification tasks. In this work, we consider extending classic SVMs with quantum kernels and applying them to satellite data analysis. The design and implementation of SVMs with quantum kernels (hybrid SVMs) is presented. It consists of the Quantum Kernel Estimation (QKE) procedure combined with a classic SVM training routine. The pixel data are mapped to the Hilbert space using ZZ-feature maps acting on the parameterized ansatz state. The parameters are optimized to maximize the kernel target alignment. We approach the problem of cloud detection in satellite image data, which is one of the pivotal steps in both on-the-ground and on-board satellite image analysis processing chains. The experiments performed over the benchmark Landsat-8 multispectral dataset revealed that the simulated hybrid SVM successfully classifies satellite images with accuracy on par with classic SVMs.

Abstract: One-shot skeleton action recognition, which aims to learn a skeleton action recognition model with a single training sample, has attracted increasing interest due to the challenge of collecting and annotating large-scale skeleton action data. However, most existing studies match skeleton sequences by comparing their feature vectors directly which neglects spatial structures and temporal orders of skeleton data. This paper presents a novel one-shot skeleton action recognition technique that handles skeleton action recognition via multi-scale spatial-temporal feature matching. We represent skeleton data at multiple spatial and temporal scales and achieve optimal feature matching from two perspectives. The first is multi-scale matching which captures the scale-wise semantic relevance of skeleton data at multiple spatial and temporal scales simultaneously. The second is cross-scale matching which handles different motion magnitudes and speeds by capturing sample-wise relevance across multiple scales. Extensive experiments over three large-scale datasets (NTU RGB+D, NTU RGB+D 120, and PKU-MMD) show that our method achieves superior one-shot skeleton action recognition, and it outperforms the state-of-the-art consistently by large margins.

Abstract: Multispectral and Hyperspectral Image Fusion (MHIF) is a practical task that aims to fuse a high-resolution multispectral image (HR-MSI) and a low-resolution hyperspectral image (LR-HSI) of the same scene to obtain a high-resolution hyperspectral image (HR-HSI). Benefiting from powerful inductive bias capability, CNN-based methods have achieved great success in the MHIF task. However, they lack certain interpretability and require convolution structures be stacked to enhance performance. Recently, Implicit Neural Representation (INR) has achieved good performance and interpretability in 2D tasks due to its ability to locally interpolate samples and utilize multimodal content such as pixels and coordinates. Although INR-based approaches show promise, they require extra construction of high-frequency information (\emph{e.g.,} positional encoding). In this paper, inspired by previous work of MHIF task, we realize that HR-MSI could serve as a high-frequency detail auxiliary input, leading us to propose a novel INR-based hyperspectral fusion function named Implicit Neural Feature Fusion Function (INF). As an elaborate structure, it solves the MHIF task and addresses deficiencies in the INR-based approaches. Specifically, our INF designs a Dual High-Frequency Fusion (DHFF) structure that obtains high-frequency information twice from HR-MSI and LR-HSI, then subtly fuses them with coordinate information. Moreover, the proposed INF incorporates a parameter-free method named INR with cosine similarity (INR-CS) that uses cosine similarity to generate local weights through feature vectors. Based on INF, we construct an Implicit Neural Fusion Network (INFN) that achieves state-of-the-art performance for MHIF tasks of two public datasets, \emph{i.e.,} CAVE and Harvard. The code will soon be made available on GitHub.

Abstract: Video Compressed Sensing (VCS) aims to reconstruct multiple frames from one single captured measurement, thus achieving high-speed scene recording with a low-frame-rate sensor. Although there have been impressive advances in VCS recently, those state-of-the-art (SOTA) methods also significantly increase model complexity and suffer from poor generality and robustness, which means that those networks need to be retrained to accommodate the new system. Such limitations hinder the real-time imaging and practical deployment of models. In this work, we propose a Sampling-Priors-Augmented Deep Unfolding Network (SPA-DUN) for efficient and robust VCS reconstruction. Under the optimization-inspired deep unfolding framework, a lightweight and efficient U-net is exploited to downsize the model while improving overall performance. Moreover, the prior knowledge from the sampling model is utilized to dynamically modulate the network features to enable single SPA-DUN to handle arbitrary sampling settings, augmenting interpretability and generality. Extensive experiments on both simulation and real datasets demonstrate that SPA-DUN is not only applicable for various sampling settings with one single model but also achieves SOTA performance with incredible efficiency.

Abstract: Myocardial infarction (MI) is one of the most prevalent cardiovascular diseases with associated clinical decision-making typically based on single-valued imaging biomarkers. However, such metrics only approximate the complex 3D structure and physiology of the heart and hence hinder a better understanding and prediction of MI outcomes. In this work, we investigate the utility of complete 3D cardiac shapes in the form of point clouds for an improved detection of MI events. To this end, we propose a fully automatic multi-step pipeline consisting of a 3D cardiac surface reconstruction step followed by a point cloud classification network. Our method utilizes recent advances in geometric deep learning on point clouds to enable direct and efficient multi-scale learning on high-resolution surface models of the cardiac anatomy. We evaluate our approach on 1068 UK Biobank subjects for the tasks of prevalent MI detection and incident MI prediction and find improvements of ~13% and ~5% respectively over clinical benchmarks. Furthermore, we analyze the role of each ventricle and cardiac phase for 3D shape-based MI detection and conduct a visual analysis of the morphological and physiological patterns typically associated with MI outcomes.

Abstract: High-resolution wide-angle fisheye images are becoming more and more important for robotics applications such as autonomous driving. However, using ordinary convolutional neural networks or vision transformers on this data is problematic due to projection and distortion losses introduced when projecting to a rectangular grid on the plane. We introduce the HEAL-SWIN transformer, which combines the highly uniform Hierarchical Equal Area iso-Latitude Pixelation (HEALPix) grid used in astrophysics and cosmology with the Hierarchical Shifted-Window (SWIN) transformer to yield an efficient and flexible model capable of training on high-resolution, distortion-free spherical data. In HEAL-SWIN, the nested structure of the HEALPix grid is used to perform the patching and windowing operations of the SWIN transformer, resulting in a one-dimensional representation of the spherical data with minimal computational overhead. We demonstrate the superior performance of our model for semantic segmentation and depth regression tasks on both synthetic and real automotive datasets. Our code is available at https://github.com/JanEGerken/HEAL-SWIN.

Abstract: In this work, we present SynTable, a unified and flexible Python-based dataset generator built using NVIDIA's Isaac Sim Replicator Composer for generating high-quality synthetic datasets for unseen object amodal instance segmentation of cluttered tabletop scenes. Our dataset generation tool can render a complex 3D scene containing object meshes, materials, textures, lighting, and backgrounds. Metadata, such as modal and amodal instance segmentation masks, occlusion masks, depth maps, bounding boxes, and material properties, can be generated to automatically annotate the scene according to the users' requirements. Our tool eliminates the need for manual labeling in the dataset generation process while ensuring the quality and accuracy of the dataset. In this work, we discuss our design goals, framework architecture, and the performance of our tool. We demonstrate the use of a sample dataset generated using SynTable by ray tracing for training a state-of-the-art model, UOAIS-Net. The results show significantly improved performance in Sim-to-Real transfer when evaluated on the OSD-Amodal dataset. We offer this tool as an open-source, easy-to-use, photorealistic dataset generator for advancing research in deep learning and synthetic data generation.

Abstract: Most image retrieval research focuses on improving predictive performance, but they may fall short in scenarios where the reliability of the prediction is crucial. Though uncertainty quantification can help by assessing uncertainty for query and database images, this method can provide only a heuristic estimate rather than an guarantee. To address these limitations, we present Risk Controlled Image Retrieval (RCIR), which generates retrieval sets that are guaranteed to contain the ground truth samples with a predefined probability. RCIR can be easily plugged into any image retrieval method, agnostic to data distribution and model selection. To the best of our knowledge, this is the first work that provides coverage guarantees for image retrieval. The validity and efficiency of RCIR is demonstrated on four real-world image retrieval datasets, including the Stanford CAR-196 (Krause et al. 2013), CUB-200 (Wah et al. 2011), the Pittsburgh dataset (Torii et al. 2013) and the ChestX-Det dataset (Lian et al. 2021).

Abstract: Most sign language translation (SLT) methods to date require the use of gloss annotations to provide additional supervision information, however, the acquisition of gloss is not easy. To solve this problem, we first perform an analysis of existing models to confirm how gloss annotations make SLT easier. We find that it can provide two aspects of information for the model, 1) it can help the model implicitly learn the location of semantic boundaries in continuous sign language videos, 2) it can help the model understand the sign language video globally. We then propose \emph{gloss attention}, which enables the model to keep its attention within video segments that have the same semantics locally, just as gloss helps existing models do. Furthermore, we transfer the knowledge of sentence-to-sentence similarity from the natural language model to our gloss attention SLT network (GASLT) to help it understand sign language videos at the sentence level. Experimental results on multiple large-scale sign language datasets show that our proposed GASLT model significantly outperforms existing methods. Our code is provided in \url{https://github.com/YinAoXiong/GASLT}.

Abstract: Computer-assisted diagnostic and prognostic systems of the future should be capable of simultaneously processing multimodal data. Multimodal deep learning (MDL), which involves the integration of multiple sources of data, such as images and text, has the potential to revolutionize the analysis and interpretation of biomedical data. However, it only caught researchers' attention recently. To this end, there is a critical need to conduct a systematic review on this topic, identify the limitations of current work, and explore future directions. In this scoping review, we aim to provide a comprehensive overview of the current state of the field and identify key concepts, types of studies, and research gaps with a focus on biomedical images and texts joint learning, mainly because these two were the most commonly available data types in MDL research. This study reviewed the current uses of multimodal deep learning on five tasks: (1) Report generation, (2) Visual question answering, (3) Cross-modal retrieval, (4) Computer-aided diagnosis, and (5) Semantic segmentation. Our results highlight the diverse applications and potential of MDL and suggest directions for future research in the field. We hope our review will facilitate the collaboration of natural language processing (NLP) and medical imaging communities and support the next generation of decision-making and computer-assisted diagnostic system development.

Abstract: Image captioning is a significant field across computer vision and natural language processing. We propose and present AIC-AB NET, a novel Attribute-Information-Combined Attention-Based Network that combines spatial attention architecture and text attributes in an encoder-decoder. For caption generation, adaptive spatial attention determines which image region best represents the image and whether to attend to the visual features or the visual sentinel. Text attribute information is synchronously fed into the decoder to help image recognition and reduce uncertainty. We have tested and evaluated our AICAB NET on the MS COCO dataset and a new proposed Fashion dataset. The Fashion dataset is employed as a benchmark of single-object images. The results show the superior performance of the proposed model compared to the state-of-the-art baseline and ablated models on both the images from MSCOCO and our single-object images. Our AIC-AB NET outperforms the baseline adaptive attention network by 0.017 (CIDEr score) on the MS COCO dataset and 0.095 (CIDEr score) on the Fashion dataset.

Abstract: The development of computer vision and in-situ monitoring using visual sensors allows the collection of large datasets from the additive manufacturing (AM) process. Such datasets could be used with machine learning techniques to improve the quality of AM. This paper examines two scenarios: first, using convolutional neural networks (CNNs) to accurately classify defects in an image dataset from AM and second, applying active learning techniques to the developed classification model. This allows the construction of a human-in-the-loop mechanism to reduce the size of the data required to train and generate training data.

Abstract: The ubiquity of camera-enabled devices has led to large amounts of unlabeled image data being produced at the edge. The integration of self-supervised learning (SSL) and federated learning (FL) into one coherent system can potentially offer data privacy guarantees while also advancing the quality and robustness of the learned visual representations without needing to move data around. However, client bias and divergence during FL aggregation caused by data heterogeneity limits the performance of learned visual representations on downstream tasks. In this paper, we propose a new aggregation strategy termed Layer-wise Divergence Aware Weight Aggregation (L-DAWA) to mitigate the influence of client bias and divergence during FL aggregation. The proposed method aggregates weights at the layer-level according to the measure of angular divergence between the clients' model and the global model. Extensive experiments with cross-silo and cross-device settings on CIFAR-10/100 and Tiny ImageNet datasets demonstrate that our methods are effective and obtain new SOTA performance on both contrastive and non-contrastive SSL approaches.

Abstract: With the growing interest in pretrained vision-language models like CLIP, recent research has focused on adapting these models to downstream tasks. Despite achieving promising results, most existing methods require labeled data for all classes, which may not hold in real-world applications due to the long tail and Zipf's law. For example, some classes may lack labeled data entirely, such as emerging concepts. To address this problem, we propose a plug-and-play generative approach called \textbf{S}ynt\textbf{H}es\textbf{I}zed \textbf{P}rompts~(\textbf{SHIP}) to improve existing fine-tuning methods. Specifically, we follow variational autoencoders to introduce a generator that reconstructs the visual features by inputting the synthesized prompts and the corresponding class names to the textual encoder of CLIP. In this manner, we easily obtain the synthesized features for the remaining label-only classes. Thereafter, we fine-tune CLIP with off-the-shelf methods by combining labeled and synthesized features. Extensive experiments on base-to-new generalization, cross-dataset transfer learning, and generalized zero-shot learning demonstrate the superiority of our approach. The code is available at \url{https://github.com/mrflogs/SHIP}.

Abstract: The Leaf Area Index (LAI) is vital for predicting winter wheat yield. Acquisition of crop conditions via Sentinel-2 remote sensing images can be hindered by persistent clouds, affecting yield predictions. Synthetic Aperture Radar (SAR) provides all-weather imagery, and the ratio between its cross- and co-polarized channels (C-band) shows a high correlation with time series LAI over winter wheat regions. This study evaluates the use of time series Sentinel-1 VH/VV for LAI imputation, aiming to increase spatial-temporal density. We utilize a bidirectional LSTM (BiLSTM) network to impute time series LAI and use half mean squared error for each time step as the loss function. We trained models on data from southern Germany and the North China Plain using only LAI data generated by Sentinel-1 VH/VV and Sentinel-2. Experimental results show BiLSTM outperforms traditional regression methods, capturing nonlinear dynamics between multiple time series. It proves robust in various growing conditions and is effective even with limited Sentinel-2 images. BiLSTM's performance surpasses that of LSTM, particularly over the senescence period. Therefore, BiLSTM can be used to impute LAI with time-series Sentinel-1 VH/VV and Sentinel-2 data, and this method could be applied to other time-series imputation issues.

Abstract: Recognizing interactive action plays an important role in human-robot interaction and collaboration. Previous methods use late fusion and co-attention mechanism to capture interactive relations, which have limited learning capability or inefficiency to adapt to more interacting entities. With assumption that priors of each entity are already known, they also lack evaluations on a more general setting addressing the diversity of subjects. To address these problems, we propose an Interactive Spatiotemporal Token Attention Network (ISTA-Net), which simultaneously model spatial, temporal, and interactive relations. Specifically, our network contains a tokenizer to partition Interactive Spatiotemporal Tokens (ISTs), which is a unified way to represent motions of multiple diverse entities. By extending the entity dimension, ISTs provide better interactive representations. To jointly learn along three dimensions in ISTs, multi-head self-attention blocks integrated with 3D convolutions are designed to capture inter-token correlations. When modeling correlations, a strict entity ordering is usually irrelevant for recognizing interactive actions. To this end, Entity Rearrangement is proposed to eliminate the orderliness in ISTs for interchangeable entities. Extensive experiments on four datasets verify the effectiveness of ISTA-Net by outperforming state-of-the-art methods. Our code is publicly available at https://github.com/Necolizer/ISTA-Net

Abstract: Whole slide image (WSI) assessment is a challenging and crucial step in cancer diagnosis and treatment planning. WSIs require high magnifications to facilitate sub-cellular analysis. Precise annotations for patch- or even pixel-level classifications in the context of gigapixel WSIs are tedious to acquire and require domain experts. Coarse-grained labels, on the other hand, are easily accessible, which makes WSI classification an ideal use case for multiple instance learning (MIL). In our work, we propose a novel embedding-based Dual-Query MIL pipeline (DQ-MIL). We contribute to both the embedding and aggregation steps. Since all-purpose visual feature representations are not yet available, embedding models are currently limited in terms of generalizability. With our work, we explore the potential of dynamic meta-embedding based on cutting-edge self-supervised pre-trained models in the context of MIL. Moreover, we propose a new MIL architecture capable of combining MIL-attention with correlated self-attention. The Dual-Query Perceiver design of our approach allows us to leverage the concept of self-distillation and to combine the advantages of a small model in the context of a low data regime with the rich feature representation of a larger model. We demonstrate the superior performance of our approach on three histopathological datasets, where we show improvement of up to 10% over state-of-the-art approaches.

Abstract: The focal point of egocentric video understanding is modelling hand-object interactions. Standard models, e.g. CNNs or Vision Transformers, which receive RGB frames as input perform well. However, their performance improves further by employing additional input modalities that provide complementary cues, such as object detections, optical flow, audio, etc. The added complexity of the modality-specific modules, on the other hand, makes these models impractical for deployment. The goal of this work is to retain the performance of such a multimodal approach, while using only the RGB frames as input at inference time. We demonstrate that for egocentric action recognition on the Epic-Kitchens and the Something-Something datasets, students which are taught by multimodal teachers tend to be more accurate and better calibrated than architecturally equivalent models trained on ground truth labels in a unimodal or multimodal fashion. We further adopt a principled multimodal knowledge distillation framework, allowing us to deal with issues which occur when applying multimodal knowledge distillation in a naive manner. Lastly, we demonstrate the achieved reduction in computational complexity, and show that our approach maintains higher performance with the reduction of the number of input views.

Abstract: In this work, we introduce a self-supervised feature representation learning framework DreamTeacher that utilizes generative networks for pre-training downstream image backbones. We propose to distill knowledge from a trained generative model into standard image backbones that have been well engineered for specific perception tasks. We investigate two types of knowledge distillation: 1) distilling learned generative features onto target image backbones as an alternative to pretraining these backbones on large labeled datasets such as ImageNet, and 2) distilling labels obtained from generative networks with task heads onto logits of target backbones. We perform extensive analyses on multiple generative models, dense prediction benchmarks, and several pre-training regimes. We empirically find that our DreamTeacher significantly outperforms existing self-supervised representation learning approaches across the board. Unsupervised ImageNet pre-training with DreamTeacher leads to significant improvements over ImageNet classification pre-training on downstream datasets, showcasing generative models, and diffusion generative models specifically, as a promising approach to representation learning on large, diverse datasets without requiring manual annotation.

Abstract: The growing threats of deepfakes to society and cybersecurity have raised enormous public concerns, and increasing efforts have been devoted to this critical topic of deepfake video detection. Existing video methods achieve good performance but are computationally intensive. This paper introduces a simple yet effective strategy named Thumbnail Layout (TALL), which transforms a video clip into a pre-defined layout to realize the preservation of spatial and temporal dependencies. Specifically, consecutive frames are masked in a fixed position in each frame to improve generalization, then resized to sub-images and rearranged into a pre-defined layout as the thumbnail. TALL is model-agnostic and extremely simple by only modifying a few lines of code. Inspired by the success of vision transformers, we incorporate TALL into Swin Transformer, forming an efficient and effective method TALL-Swin. Extensive experiments on intra-dataset and cross-dataset validate the validity and superiority of TALL and SOTA TALL-Swin. TALL-Swin achieves 90.79$\%$ AUC on the challenging cross-dataset task, FaceForensics++ $\to$ Celeb-DF. The code is available at https://github.com/rainy-xu/TALL4Deepfake.

Abstract: We address the problem of generating realistic 3D motions of humans interacting with objects in a scene. Our key idea is to create a neural interaction field attached to a specific object, which outputs the distance to the valid interaction manifold given a human pose as input. This interaction field guides the sampling of an object-conditioned human motion diffusion model, so as to encourage plausible contacts and affordance semantics. To support interactions with scarcely available data, we propose an automated synthetic data pipeline. For this, we seed a pre-trained motion model, which has priors for the basics of human movement, with interaction-specific anchor poses extracted from limited motion capture data. Using our guided diffusion model trained on generated synthetic data, we synthesize realistic motions for sitting and lifting with several objects, outperforming alternative approaches in terms of motion quality and successful action completion. We call our framework NIFTY: Neural Interaction Fields for Trajectory sYnthesis.