Yang Liu

Jiayu Huo, Yang Liu, Xi Ouyang et al.

Accurately segmenting brain lesions in MRI scans is critical for providing patients with prognoses and neurological monitoring. However, the performance of CNN-based segmentation methods is constrained by the limited training set size. Advanced data augmentation is an effective strategy to improve the model's robustness. However, they often introduce intensity disparities between foreground and background areas and boundary artifacts, which weakens the effectiveness of such strategies. In this paper, we propose a foreground harmonization framework (ARHNet) to tackle intensity disparities and make synthetic images look more realistic. In particular, we propose an Adaptive Region Harmonization (ARH) module to dynamically align foreground feature maps to the background with an attention mechanism. We demonstrate the efficacy of our method in improving the segmentation performance using real and synthetic images. Experimental results on the ATLAS 2.0 dataset show that ARHNet outperforms other methods for image harmonization tasks, and boosts the down-stream segmentation performance. Our code is publicly available at https://github.com/King-HAW/ARHNet.

Rui Huang, Ruofei Wang, Qing Guo et al.

Change detection (CD) is to decouple object changes (i.e., object missing or appearing) from background changes (i.e., environment variations) like light and season variations in two images captured in the same scene over a long time span, presenting critical applications in disaster management, urban development, etc. In particular, the endless patterns of background changes require detectors to have a high generalization against unseen environment variations, making this task significantly challenging. Recent deep learning-based methods develop novel network architectures or optimization strategies with paired-training examples, which do not handle the generalization issue explicitly and require huge manual pixel-level annotation efforts. In this work, for the first attempt in the CD community, we study the generalization issue of CD from the perspective of data augmentation and develop a novel weakly supervised training algorithm that only needs image-level labels. Different from general augmentation techniques for classification, we propose the background-mixed augmentation that is specifically designed for change detection by augmenting examples under the guidance of a set of background-changing images and letting deep CD models see diverse environment variations. Moreover, we propose the augmented & real data consistency loss that encourages the generalization increase significantly. Our method as a general framework can enhance a wide range of existing deep learning-based detectors. We conduct extensive experiments in two public datasets and enhance four state-of-the-art methods, demonstrating the advantages of our method. We release the code at https://github.com/tsingqguo/bgmix.

Yang Liu, Yao Zhang, Yixin Wang et al.

Recently, the dominant DETR-based approaches apply central-concept spatial prior to accelerate Transformer detector convergency. These methods gradually refine the reference points to the center of target objects and imbue object queries with the updated central reference information for spatially conditional attention. However, centralizing reference points may severely deteriorate queries' saliency and confuse detectors due to the indiscriminative spatial prior. To bridge the gap between the reference points of salient queries and Transformer detectors, we propose SAlient Point-based DETR (SAP-DETR) by treating object detection as a transformation from salient points to instance objects. In SAP-DETR, we explicitly initialize a query-specific reference point for each object query, gradually aggregate them into an instance object, and then predict the distance from each side of the bounding box to these points. By rapidly attending to query-specific reference region and other conditional extreme regions from the image features, SAP-DETR can effectively bridge the gap between the salient point and the query-based Transformer detector with a significant convergency speed. Our extensive experiments have demonstrated that SAP-DETR achieves 1.4 times convergency speed with competitive performance. Under the standard training scheme, SAP-DETR stably promotes the SOTA approaches by 1.0 AP. Based on ResNet-DC-101, SAP-DETR achieves 46.9 AP.

Farnoosh Javadi, Walid Ahmed, Habib Hajimolahoseini et al.

Massive transformer-based models face several challenges, including slow and computationally intensive pre-training and over-parametrization. This paper addresses these challenges by proposing a versatile method called GQKVA, which generalizes query, key, and value grouping techniques. GQKVA is designed to speed up transformer pre-training while reducing the model size. Our experiments with various GQKVA variants highlight a clear trade-off between performance and model size, allowing for customized choices based on resource and time limitations. Our findings also indicate that the conventional multi-head attention approach is not always the best choice, as there are lighter and faster alternatives available. We tested our method on ViT, which achieved an approximate 0.3% increase in accuracy while reducing the model size by about 4% in the task of image classification. Additionally, our most aggressive model reduction experiment resulted in a reduction of approximately 15% in model size, with only around a 1% drop in accuracy.

Yang Liu, Yao Zhang, Yixin Wang et al.

Transformer, an attention-based encoder-decoder model, has already revolutionized the field of natural language processing (NLP). Inspired by such significant achievements, some pioneering works have recently been done on employing Transformer-liked architectures in the computer vision (CV) field, which have demonstrated their effectiveness on three fundamental CV tasks (classification, detection, and segmentation) as well as multiple sensory data stream (images, point clouds, and vision-language data). Because of their competitive modeling capabilities, the visual Transformers have achieved impressive performance improvements over multiple benchmarks as compared with modern Convolution Neural Networks (CNNs). In this survey, we have reviewed over one hundred of different visual Transformers comprehensively according to three fundamental CV tasks and different data stream types, where a taxonomy is proposed to organize the representative methods according to their motivations, structures, and application scenarios. Because of their differences on training settings and dedicated vision tasks, we have also evaluated and compared all these existing visual Transformers under different configurations. Furthermore, we have revealed a series of essential but unexploited aspects that may empower such visual Transformers to stand out from numerous architectures, e.g., slack high-level semantic embeddings to bridge the gap between the visual Transformers and the sequential ones. Finally, three promising research directions are suggested for future investment. We will continue to update the latest articles and their released source codes at https://github.com/liuyang-ict/awesome-visual-transformers.

Tu Zheng, Shuai Zhao, Yang Liu et al.

Bounding box regression is an important component in object detection. Recent work achieves promising performance by optimizing the Intersection over Union~(IoU). However, IoU-based loss has the gradient vanish problem in the case of low overlapping bounding boxes, and the model could easily ignore these simple cases. In this paper, we propose Side Overlap~(SO) loss by maximizing the side overlap of two bounding boxes, which puts more penalty for low overlapping bounding box cases. Besides, to speed up the convergence, the Corner Distance~(CD) is added into the objective function. Combining the Side Overlap and Corner Distance, we get a new regression objective function, \textit{Side and Corner Align Loss~(SCALoss)}. The SCALoss is well-correlated with IoU loss, which also benefits the evaluation metric but produces more penalty for low-overlapping cases. It can serve as a comprehensive similarity measure, leading to better localization performance and faster convergence speed. Experiments on COCO, PASCAL VOC, and LVIS benchmarks show that SCALoss can bring consistent improvement and outperform $\ell_n$ loss and IoU based loss with popular object detectors such as YOLOV3, SSD, Faster-RCNN. Code is available at: \url{https://github.com/Turoad/SCALoss}.

Foozhan Ataiefard, Walid Ahmed, Habib Hajimolahoseini et al.

Vision transformers are known to be more computationally and data-intensive than CNN models. These transformer models such as ViT, require all the input image tokens to learn the relationship among them. However, many of these tokens are not informative and may contain irrelevant information such as unrelated background or unimportant scenery. These tokens are overlooked by the multi-head self-attention (MHSA), resulting in many redundant and unnecessary computations in MHSA and the feed-forward network (FFN). In this work, we propose a method to optimize the amount of unnecessary interactions between unimportant tokens by separating and sending them through a different low-cost computational path. Our method does not add any parameters to the ViT model and aims to find the best trade-off between training throughput and achieving a 0% loss in the Top-1 accuracy of the final model. Our experimental results on training ViT-small from scratch show that SkipViT is capable of effectively dropping 55% of the tokens while gaining more than 13% training throughput and maintaining classification accuracy at the level of the baseline model on Huawei Ascend910A.

Yu-Qi Yang, Peng-Shuai Wang, Yang Liu

Sparse voxel-based 3D convolutional neural networks (CNNs) are widely used for various 3D vision tasks. Sparse voxel-based 3D CNNs create sparse non-empty voxels from the 3D input and perform 3D convolution operations on them only. We propose a simple yet effective padding scheme --- interpolation-aware padding to pad a few empty voxels adjacent to the non-empty voxels and involve them in the 3D CNN computation so that all neighboring voxels exist when computing point-wise features via the trilinear interpolation. For fine-grained 3D vision tasks where point-wise features are essential, like semantic segmentation and 3D detection, our network achieves higher prediction accuracy than the existing networks using the nearest neighbor interpolation or the normalized trilinear interpolation with the zero-padding or the octree-padding scheme. Through extensive comparisons on various 3D segmentation and detection tasks, we demonstrate the superiority of 3D sparse CNNs with our padding scheme in conjunction with feature interpolation.

Xinle Liang, Yang Liu, Jiahuan Luo et al.

Federated Learning (FL) provides both model performance and data privacy for machine learning tasks where samples or features are distributed among different parties. In the training process of FL, no party has a global view of data distributions or model architectures of other parties. Thus the manually-designed architectures may not be optimal. In the past, Neural Architecture Search (NAS) has been applied to FL to address this critical issue. However, existing Federated NAS approaches require prohibitive communication and computation effort, as well as the availability of high-quality labels. In this work, we present Self-supervised Vertical Federated Neural Architecture Search (SS-VFNAS) for automating FL where participants hold feature-partitioned data, a common cross-silo scenario called Vertical Federated Learning (VFL). In the proposed framework, each party first conducts NAS using self-supervised approach to find a local optimal architecture with its own data. Then, parties collaboratively improve the local optimal architecture in a VFL framework with supervision. We demonstrate experimentally that our approach has superior performance, communication efficiency and privacy compared to Federated NAS and is capable of generating high-performance and highly-transferable heterogeneous architectures even with insufficient overlapping samples, providing automation for those parties without deep learning expertise.

Yixin Wang, Yao Zhang, Feng Hou et al.

Automatic brain tumor segmentation from multi-modality Magnetic Resonance Images (MRI) using deep learning methods plays an important role in assisting the diagnosis and treatment of brain tumor. However, previous methods mostly ignore the latent relationship among different modalities. In this work, we propose a novel end-to-end Modality-Pairing learning method for brain tumor segmentation. Paralleled branches are designed to exploit different modality features and a series of layer connections are utilized to capture complex relationships and abundant information among modalities. We also use a consistency loss to minimize the prediction variance between two branches. Besides, learning rate warmup strategy is adopted to solve the problem of the training instability and early over-fitting. Lastly, we use average ensemble of multiple models and some post-processing techniques to get final results. Our method is tested on the BraTS 2020 online testing dataset, obtaining promising segmentation performance, with average dice scores of 0.891, 0.842, 0.816 for the whole tumor, tumor core and enhancing tumor, respectively. We won the second place of the BraTS 2020 Challenge for the tumor segmentation task.