Sungnyun Kim, Haofu Liao, Srikar Appalaraju, Peng Tang, Zhuowen Tu, Ravi Kumar Satzoda, R. Manmatha, Vijay Mahadevan, Stefano Soatto. Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing. 2024.

We present a new formulation for structured information extraction (SIE) from visually rich documents. We address the limitations of existing IOB tagging and graph-based formulations, which are either overly reliant on the correct ordering of input text or struggle with decoding a complex graph. Instead, motivated by anchor-based object detectors in computer vision, we represent an entity as an anchor word and a bounding box, and represent entity linking as the association between anchor words. This is more robust to text ordering, and maintains a compact graph for entity linking. The formulation motivates us to introduce 1) a Document Transformer (DocTr) that aims at detecting and associating entity bounding boxes in visually rich documents, and 2) a simple pre-training strategy that helps learn entity detection in the context of language. Evaluations on three SIE benchmarks show the effectiveness of the proposed formulation, and the overall approach outperforms existing solutions.

A radiograph visualizes the internal anatomy of a patient through the use of X-ray, which projects 3D information onto a 2D plane. Hence, radiograph analysis naturally requires physicians to relate their prior knowledge about 3D human anatomy to 2D radiographs. Synthesizing novel radiographic views in a small range can assist physicians in interpreting anatomy more reliably; however, radiograph view synthesis is heavily ill-posed, lacking in paired data, and lacking in differentiable operations to leverage learning-based approaches. To address these problems, we use Computed Tomography (CT) for radiograph simulation and design a differentiable projection algorithm, which enables us to achieve geometrically consistent transformations between the radiography and CT domains. Our method, XraySyn, can synthesize novel views on real radiographs through a combination of realistic simulation and finetuning on real radiographs. To the best of our knowledge, this is the first work on radiograph view synthesis. We show that by gaining an understanding of radiography in 3D space, our method can be applied to radiograph bone extraction and suppression without requiring groundtruth bone labels.

Deep learning algorithms mine knowledge from the training data and thus would likely inherit the dataset’s bias information. As a result, the obtained model would generalize poorly and even mislead the decision process in real-life applications. We propose to remove the bias information misused by the target task with a crosssample adversarial debiasing (CSAD) method. CSAD explicitly extracts target and bias features disentangled from the latent representation generated by a feature extractor and then learns to discover and remove the correlation between the target and bias features. The correlation measurement plays a critical role in adversarial debiasing and is conducted by a cross-sample neural mutual information estimator. Moreover, we propose joint content and local structural representation learning to boost mutual information estimation for better performance. We conduct thorough experiments on publicly available datasets to validate the advantages of the proposed method over state-of-the-art approaches.

Computer vision applications such as visual relationship detection and human object interaction can be formulated as a composite (structured) set detection problem in which both the parts (subject, object, and predicate) and the sum (triplet as a whole) are to be detected in a hierarchical fashion. In this paper, we present a new approach, denoted Part-and-Sum detection Transformer (PST), to perform end-to-end visual composite set detection. Different from existing Transformers in which queries are at a single level, we simultaneously model the joint part and sum hypotheses/interactions with composite queries and attention modules. We explicitly incorporate sum queries to enable better modeling of the part-and-sum relations that are absent in the standard Transformers. Our approach also uses novel tensor-based part queries and vector-based sum queries, and models their joint interaction. We report experiments on two vision tasks, visual relationship detection and human object interaction and demonstrate that PST achieves state of the art results among single-stage models, while nearly matching the results of custom designed two-stage models.

Deep learning-based single image super-resolution (SISR) methods face various challenges when applied to 3D medical volumetric data (i.e., CT and MR images) due to the high memory cost and anisotropic resolution, which adversely affect their performance. Furthermore, mainstream SISR methods are designed to work over specific upsampling factors, which makes them ineffective in clinical practice. In this paper, we introduce a Spatially Aware Interpolation NeTwork (SAINT) for medical slice synthesis to alleviate the memory constraint that volumetric data poses. Compared to other super-resolution methods, SAINT utilizes voxel spacing information to provide desirable levels of details, and allows for the upsampling factor to be determined on the fly. Our evaluations based on 853 CT scans from four datasets that contain liver, colon, hepatic vessels, and kidneys show that SAINT consistently outperforms other SISR methods in terms of medical slice synthesis quality, while using only a single model to deal with different upsampling factors.

We present a novel unsupervised learning approach to image landmark discovery by incorporating the inter-subject landmark consistencies on facial images. This is achieved via an inter-subject mapping module that transforms original subject landmarks based on an auxiliary subject-related structure. To recover from the transformed images back to the original subject, the landmark detector is forced to learn spatial locations that contain the consistent semantic meanings both for the paired intra-subject images and between the paired inter-subject images. Our proposed method is extensively evaluated on two public facial image datasets (MAFL, AFLW) with various settings. Experimental results indicate that our method can extract the consistent landmarks for both datasets and achieve better performances compared to the previous state-of-the-art methods quantitatively and qualitatively.

Current deep neural network based approaches to computed tomography (CT) metal artifact reduction (MAR) are supervised methods that rely on synthesized metal artifacts for training. However, as synthesized data may not accurately simulate the underlying physical mechanisms of CT imaging, the supervised methods often generalize poorly to clinical applications. To address this problem, we propose, to the best of our knowledge, the first unsupervised learning approach to MAR. Specifically, we introduce a novel artifact disentanglement network that disentangles the metal artifacts from CT images in the latent space. It supports different forms of generations (artifact reduction, artifact transfer, and self-reconstruction, etc.) with specialized loss functions to obviate the need for supervision with synthesized data. Extensive experiments show that when applied to a synthesized dataset, our method addresses metal artifacts significantly better than the existing unsupervised models designed for natural image-to-image translation problems, and achieves comparable performance to existing supervised models for MAR. When applied to clinical datasets, our method demonstrates better generalization ability over the supervised models. The source code of this paper is publicly available at https:// github.com/liaohaofu/adn.

We propose to tackle the problem of multiview 2D/3D rigid registration for intervention via a Point-Of-Interest Network for Tracking and Triangulation (POINT2). POINT2learns to establish 2D point-to-point correspondences between the pre- and intra-intervention images by tracking a set of random POIs. The 3D pose of the pre-intervention volume is then estimated through a triangulation layer. In POINT2, the unified framework of the POI tracker and the triangulation layer enables learning informative 2D features and estimating 3D pose jointly. In contrast to existing approaches, POINT2only requires a single forward-pass to achieve a reliable 2D/3D registration. As the POI tracker is shift-invariant, POINT2is more robust to the initial pose of the 3D pre-intervention image. Extensive experiments on a large-scale clinical cone-beam CT (CBCT) dataset show that the proposed POINT2method outperforms the existing learning-based method in terms of accuracy, robustness and running time. Furthermore, when used as an initial pose estimator, our method also improves the robustness and speed of the state-of-the-art optimization-based approaches by ten folds.

Computed tomography (CT) is an imaging modality widely used for medical diagnosis and treatment. CT images are often corrupted by undesirable artifacts when metallic implants are carried by patients, which creates the problem of metal artifact reduction (MAR). Existing methods for reducing the artifacts due to metallic implants are inadequate for two main reasons. First, metal artifacts are structured and non-local so that simple image domain enhancement approaches would not suffice. Second, the MAR approaches which attempt to reduce metal artifacts in the X-ray projection (sinogram) domain inevitably lead to severe secondary artifact due to sinogram inconsistency. To overcome these difficulties, we propose an end-to-end trainable Dual Domain Network (DuDoNet) to simultaneously restore sinogram consistency and enhance CT images. The linkage between the sigogram and image domains is a novel Radon inversion layer that allows the gradients to back-propagate from the image domain to the sinogram domain during training. Extensive experiments show that our method achieves significant improvements over other single domain MAR approaches. To the best of our knowledge, it is the first end-to-end dual-domain network for MAR.

Automatic vertebrae identification and localization from arbitrary computed tomography (CT) images is challenging. Vertebrae usually share similar morphological appearance. Because of pathology and the arbitrary field-of-view of CT scans, one can hardly rely on the existence of some anchor vertebrae or parametric methods to model the appearance and shape. To solve the problem, we argue that: 1) one should make use of the short-range contextual information, such as the presence of some nearby organs (if any), to roughly estimate the target vertebrae; and 2) due to the unique anatomic structure of the spine column, vertebrae have fixed sequential order, which provides the important long-range contextual information to further calibrate the results. We propose a robust and efficient vertebrae identification and localization system that can inherently learn to incorporate both the short- and long-range contextual information in a supervised manner. To this end, we develop a multi-task 3-D fully convolutional neural network to effectively extract the short-range contextual information around the target vertebrae. For the long-range contextual information, we propose a multi-task bidirectional recurrent neural network to encode the spatial and contextual information among the vertebrae of the visible spine column. We demonstrate the effectiveness of the proposed approach on a challenging data set, and the experimental results show that our approach outperforms the state-of-the-art methods by a significant margin.

When looking for a restaurant online, user uploaded photos often give people an immediate and tangible impression about a restaurant. Due to their informativeness, such user contributed photos are leveraged by restaurant review websites to provide their users an intuitive and effective search experience. In this paper, we present a novel approach to inferring restaurant types or styles (ambiance, dish styles, suitability for different occasions) from user uploaded photos on user-review websites. To that end, we first collect a novel restaurant photo dataset associating the user contributed photos with the restaurant styles from TripAdvior. We then propose a deep multi-instance multi-label learning (MIML) framework to deal with the unique problem setting of the restaurant style classification task. We employ a two-step bootstrap strategy to train a multi-label convolutional neural network (CNN). The multi-label CNN is then used to compute the confidence scores of restaurant styles for all the images associated with a restaurant. The computed confidence scores are further used to train a final binary classifier for each restaurant style tag. Upon training, the styles of a restaurant can be profiled by analyzing restaurant photos with the trained multi-label CNN and SVM models. Experimental evaluation has demonstrated that our crowd sourcing-based approach can effectively infer the restaurant style when there are a sufficient number of user uploaded photos for a given restaurant.

In this study, we investigate what a practically useful approach is in order to achieve robust skin disease diagnosis. A direct approach is to target the ground truth diagnosis labels, while an alternative approach instead focuses on determining skin lesion characteristics that are more visually consistent and discernible. We argue that, for computer aided skin disease diagnosis, it is both more realistic and more useful that lesion type tags should be considered as the target of an automated diagnosis system such that the system can first achieve a high accuracy in describing skin lesions, and in turn facilitate disease diagnosis using lesion characteristics in conjunction with other evidences. To further meet such an objective, we employ convolutional neutral networks (CNNs) for both the disease-targeted and lesion-targeted classifications. We have collected a large-scale and diverse dataset of 75,665 skin disease images from six publicly available dermatology atlantes. Then we train and compare both disease-targeted and lesion-targeted classifiers, respectively. For disease-targeted classification, only 27.6% top-1 accuracy and 57.9% top-5 accuracy are achieved with a mean average precision (mAP) of 0.42. In contrast, for lesion-targeted classification, we can achieve a much higher mAP of 0.70.