Clinical classification of chest radiography is particularly challenging for standard machine learning algorithms due to its inherent long-tailed and multi-label nature. However, few attempts take into account the coupled challenges posed by both the class imbalance and label co-occurrence, which hinders their value to boost the diagnosis on chest X-rays (CXRs) in the real-world scenarios. Besides, with the prevalence of pretraining techniques, how to incorporate these new paradigms into the current framework lacks of the systematical study. This technical report presents a brief description of our solution in the ICCV CVAMD 2023 CXR-LT Competition. We empirically explored the effectiveness for CXR diagnosis with the integration of several advanced designs about data augmentation, feature extractor, classifier design, loss function reweighting, exogenous data replenishment, etc. In addition, we improve the performance through simple test-time data augmentation and ensemble. Our framework finally achieves 0.349 mAP on the competition test set, ranking in the top five.
In this study, we aim to initiate the development of Radiology Foundation Model, termed as RadFM.We consider the construction of foundational models from the perspectives of data, model design, and evaluation thoroughly. Our contribution can be concluded as follows: (i), we construct a large-scale Medical Multi-modal Dataset, MedMD, consisting of 16M 2D and 3D medical scans. To the best of our knowledge, this is the first multi-modal dataset containing 3D medical scans. (ii), We propose an architecture that enables visually conditioned generative pre-training, allowing for the integration of text input interleaved with 2D or 3D medical scans to generate response for diverse radiologic tasks. The model was initially pre-trained on MedMD and subsequently domain-specific fine-tuned on RadMD, a radiologic cleaned version of MedMD, containing 3M radiologic visual-language pairs. (iii), we propose a new evaluation benchmark that comprises five tasks, aiming to comprehensively assess the capability of foundation models in handling practical clinical problems. Our experimental results confirm that RadFM significantly outperforms existing multi-modal foundation models. The codes, data, and model checkpoint will all be made publicly available to promote further research and development in the field.
Multi-person motion prediction is a challenging problem due to the dependency of motion on both individual past movements and interactions with other people. Transformer-based methods have shown promising results on this task, but they miss the explicit relation representation between joints, such as skeleton structure and pairwise distance, which is crucial for accurate interaction modeling. In this paper, we propose the Joint-Relation Transformer, which utilizes relation information to enhance interaction modeling and improve future motion prediction. Our relation information contains the relative distance and the intra-/inter-person physical constraints. To fuse relation and joint information, we design a novel joint-relation fusion layer with relation-aware attention to update both features. Additionally, we supervise the relation information by forecasting future distance. Experiments show that our method achieves a 13.4% improvement of 900ms VIM on 3DPW-SoMoF/RC and 17.8%/12.0% improvement of 3s MPJPE on CMU-Mpcap/MuPoTS-3D dataset.
Anomaly detection has gained considerable attention due to its broad range of applications, particularly in industrial defect detection. To address the challenges of data collection, researchers have introduced zero-/few-shot anomaly detection techniques that require minimal normal images for each category. However, complex industrial scenarios often involve multiple objects, presenting a significant challenge. In light of this, we propose a straightforward yet powerful multi-scale memory comparison framework for zero-/few-shot anomaly detection. Our approach employs a global memory bank to capture features across the entire image, while an individual memory bank focuses on simplified scenes containing a single object. The efficacy of our method is validated by its remarkable achievement of 4th place in the zero-shot track and 2nd place in the few-shot track of the Visual Anomaly and Novelty Detection (VAND) competition.
Class incremental learning (CIL) aims to incrementally update a trained model with the new classes of samples (plasticity) while retaining previously learned ability (stability). To address the most challenging issue in this goal, i.e., catastrophic forgetting, the mainstream paradigm is memory-replay CIL, which consolidates old knowledge by replaying a small number of old classes of samples saved in the memory. Despite effectiveness, the inherent destruction-reconstruction dynamics in memory-replay CIL are an intrinsic limitation: if the old knowledge is severely destructed, it will be quite hard to reconstruct the lossless counterpart. Our theoretical analysis shows that the destruction of old knowledge can be effectively alleviated by balancing the contribution of samples from the current phase and those saved in the memory. Motivated by this theoretical finding, we propose a novel Balanced Destruction-Reconstruction module (BDR) for memory-replay CIL, which can achieve better knowledge reconstruction by reducing the degree of maximal destruction of old knowledge. Specifically, to achieve a better balance between old knowledge and new classes, the proposed BDR module takes into account two factors: the variance in training status across different classes and the quantity imbalance of samples from the current phase and memory. By dynamically manipulating the gradient during training based on these factors, BDR can effectively alleviate knowledge destruction and improve knowledge reconstruction. Extensive experiments on a range of CIL benchmarks have shown that as a lightweight plug-and-play module, BDR can significantly improve the performance of existing state-of-the-art methods with good generalization.
Electrocardiogram (ECG) is a widely used diagnostic tool for detecting heart conditions. Rare cardiac diseases may be underdiagnosed using traditional ECG analysis, considering that no training dataset can exhaust all possible cardiac disorders. This paper proposes using anomaly detection to identify any unhealthy status, with normal ECGs solely for training. However, detecting anomalies in ECG can be challenging due to significant inter-individual differences and anomalies present in both global rhythm and local morphology. To address this challenge, this paper introduces a novel multi-scale cross-restoration framework for ECG anomaly detection and localization that considers both local and global ECG characteristics. The proposed framework employs a two-branch autoencoder to facilitate multi-scale feature learning through a masking and restoration process, with one branch focusing on global features from the entire ECG and the other on local features from heartbeat-level details, mimicking the diagnostic process of cardiologists. Anomalies are identified by their high restoration errors. To evaluate the performance on a large number of individuals, this paper introduces a new challenging benchmark with signal point-level ground truths annotated by experienced cardiologists. The proposed method demonstrates state-of-the-art performance on this benchmark and two other well-known ECG datasets. The benchmark dataset and source code are available at: \url{https://github.com/MediaBrain-SJTU/ECGAD}
The goal of the audio-visual segmentation (AVS) task is to segment the sounding objects in the video frames using audio cues. However, current fusion-based methods have the performance limitations due to the small receptive field of convolution and inadequate fusion of audio-visual features. To overcome these issues, we propose a novel \textbf{Au}dio-aware query-enhanced \textbf{TR}ansformer (AuTR) to tackle the task. Unlike existing methods, our approach introduces a multimodal transformer architecture that enables deep fusion and aggregation of audio-visual features. Furthermore, we devise an audio-aware query-enhanced transformer decoder that explicitly helps the model focus on the segmentation of the pinpointed sounding objects based on audio signals, while disregarding silent yet salient objects. Experimental results show that our method outperforms previous methods and demonstrates better generalization ability in multi-sound and open-set scenarios.
Current mainstream vision-language (VL) tracking framework consists of three parts, \ie a visual feature extractor, a language feature extractor, and a fusion model. To pursue better performance, a natural modus operandi for VL tracking is employing customized and heavier unimodal encoders, and multi-modal fusion models. Albeit effective, existing VL trackers separate feature extraction and feature integration, resulting in extracted features that lack semantic guidance and have limited target-aware capability in complex scenarios, \eg similar distractors and extreme illumination. In this work, inspired by the recent success of exploring foundation models with unified architecture for both natural language and computer vision tasks, we propose an All-in-One framework, which learns joint feature extraction and interaction by adopting a unified transformer backbone. Specifically, we mix raw vision and language signals to generate language-injected vision tokens, which we then concatenate before feeding into the unified backbone architecture. This approach achieves feature integration in a unified backbone, removing the need for carefully-designed fusion modules and resulting in a more effective and efficient VL tracking framework. To further improve the learning efficiency, we introduce a multi-modal alignment module based on cross-modal and intra-modal contrastive objectives, providing more reasonable representations for the unified All-in-One transformer backbone. Extensive experiments on five benchmarks, \ie OTB99-L, TNL2K, LaSOT, LaSOT$_{\rm Ext}$ and WebUAV-3M, demonstrate the superiority of the proposed tracker against existing state-of-the-arts on VL tracking. Codes will be made publicly available.
In semantic segmentation, adapting a visual system to novel object categories at inference time has always been both valuable and challenging. To enable such generalization, existing methods rely on either providing several support examples as visual cues or class names as textual cues. Through the development is relatively optimistic, these two lines have been studied in isolation, neglecting the complementary intrinsic of low-level visual and high-level language information. In this paper, we define a unified setting termed as open-set semantic segmentation (O3S), which aims to learn seen and unseen semantics from both visual examples and textual names. Our pipeline extracts multi-modal prototypes for segmentation task, by first single modal self-enhancement and aggregation, then multi-modal complementary fusion. To be specific, we aggregate visual features into several tokens as visual prototypes, and enhance the class name with detailed descriptions for textual prototype generation. The two modalities are then fused to generate multi-modal prototypes for final segmentation. On both \pascal and \coco datasets, we conduct extensive experiments to evaluate the framework effectiveness. State-of-the-art results are achieved even on more detailed part-segmentation, Pascal-Animals, by only training on coarse-grained datasets. Thorough ablation studies are performed to dissect each component, both quantitatively and qualitatively.