PSAQ-ViT V2: Toward Accurate and General Data-Free Quantization for Vision Transformers.

Data-free quantization can potentially address data privacy and security concerns in model compression and thus has been widely investigated. Recently, patch similarity aware data-free quantization for vision transformers (PSAQ-ViT) designs a relative value metric, patch similarity, to generate data from pretrained vision transformers (ViTs), achieving the first attempt at data-free quantization for ViTs. In this article, we propose PSAQ-ViT V2, a more accurate and general data-free quantization framework for ViTs, built on top of PSAQ-ViT. More specifically, following the patch similarity metric in PSAQ-ViT, we introduce an adaptive teacher-student strategy, which facilitates the constant cyclic evolution of the generated samples and the quantized model in a competitive and interactive fashion under the supervision of the full-precision (FP) model (teacher), thus significantly improving the accuracy of the quantized model. Moreover, without the auxiliary category guidance, we employ the task-and model-independent prior information, making the general-purpose scheme compatible with a broad range of vision tasks and models. Extensive experiments are conducted on various models on image classification, object detection, and semantic segmentation tasks, and PSAQ-ViT V2, with the naive quantization strategy and without access to real-world data, consistently achieves competitive results, showing potential as a powerful baseline on data-free quantization for ViTs. For instance, with Swin-S as the (backbone) model, 8-bit quantization reaches 82.13 top-1 accuracy on ImageNet, 50.9 box AP and 44.1 mask AP on COCO, and 47.2 mean Intersection over Union (mIoU) on ADE20K. We hope that accurate and general PSAQ-ViT V2 can serve as a potential and practice solution in real-world applications involving sensitive data. Code is released and merged at: https://github.com/zkkli/PSAQ-ViT.

Estimation of coronary artery movement using a non-rigid registration with global-local structure preservation.

BACKGROUND: At present, coronary artery disease (CAD) is the leading cause of death worldwide. Many studies have shown that CAD is strongly associated with the motion characteristics of the coronary arteries. Although cardiovascular imaging technology has been widely used for the diagnosis of CAD, the motion parameters of the heart and coronary arteries cannot be directly calculated from the images. In this paper, we propose a point set registration method with global and local topology constraints to quantify coronary artery movement. METHODS: The global constraint is the motion coherence of the point set which enforces the smoothness of the displacement field. The local linear embedding based topological structure and the local feature descriptor i.e., the 3D shape context, are designed to retain the local structure of the point set. We incorporate these constraints into a maximum likelihood framework and derive an expectation-maximization algorithm to obtain the transformation function between the two point sets. The proposed method was compared with four existing algorithms using simulated data and applied to the real data obtained from 4D CT angiograms. RESULTS: For the simulation data, the proposed method achieves a lower registration error than the comparison algorithms. For the real data, the proposed method shows that, in most cases, the right coronary artery achieves a larger velocity than the left anterior descending and left circumflex branches, and there are three well-defined velocity peaks, during the cardiac cycle for these branches. CONCLUSION: The proposed approach is feasible and effective in quantifying coronary artery movement and thus adds to the diagnostic power of coronary imaging.

Intravenous leiomyomatosis with right atrium extension in two patients: A case report.

Intravenous leiomyomatosis (IVL) is a rare benign tumor of the uterus mesoderm, which can spread via the vein, from the micro-vein to the inferior vena cava, and even to the right atrium. IVL was first described this disease in 1896 and the first presented case of IVL with right atrium extension in the autopsy was described in 1907. On rare occasions, the tumors may extend into the regional and systemic veins, thus reaching the heart. This may subsequently cause intracardiac leiomyomatosis (ICL), which may lead to congestive heart failure and on occasion, sudden fatalities. Due to its rarity and diffuse symptoms, the misdiagnosis of ICL is common and as a result, the condition may be under-reported. The present study reported two cases of IVL resected at the General Hospital of Shenyang Military Command. A hysteromyomectomy for uterine fibroids was performed on each patient (Case 1, 41 years ago; Case 2, 3 years ago). One patient presented with chest pain following a period of activity and the other presented with heart neoplasm during a routine health examination.