Large-scale Pre-Training Vision-Language Model such as CLIP has demonstrated outstanding performance in zero-shot classification, e.g. achieving 76.3% top-1 accuracy on ImageNet without seeing any example, which leads to potential benefits to many tasks that have no labeled data. However, while applying CLIP to a downstream target domain, the presence of visual and text domain gaps and cross-modality misalignment can greatly impact the model performance. To address such challenges, we propose ReCLIP, the first source-free domain adaptation method for vision-language models, which does not require any source data or target labeled data. ReCLIP first learns a projection space to mitigate the misaligned visual-text embeddings and learns pseudo labels, and then deploys cross-modality self-training with the pseudo labels, to update visual and text encoders, refine labels and reduce domain gaps and misalignments iteratively. With extensive experiments, we demonstrate ReCLIP reduces the average error rate of CLIP from 30.17% to 25.06% on 22 image classification benchmarks.
The Pulmonary Function Test (PFT) is an widely utilized and rigorous classification test for lung function evaluation, serving as a comprehensive tool for lung diagnosis. Meanwhile, Electrical Impedance Tomography (EIT) is a rapidly advancing clinical technique that visualizes conductivity distribution induced by ventilation. EIT provides additional spatial and temporal information on lung ventilation beyond traditional PFT. However, relying solely on conventional isolated interpretations of PFT results and EIT images overlooks the continuous dynamic aspects of lung ventilation. This study aims to classify lung ventilation patterns by extracting spatial and temporal features from the 3D EIT image series. The study uses a Variational Autoencoder network with a MultiRes block to compress the spatial distribution in a 3D image into a one-dimensional vector. These vectors are then concatenated to create a feature map for the exhibition of temporal features. A simple convolutional neural network is used for classification. Data collected from 137 subjects were finally used for training. The model is validated by ten-fold and leave-one-out cross-validation first. The accuracy and sensitivity of normal ventilation mode are 0.95 and 1.00, and the f1-score is 0.94. Furthermore, we check the reliability and feasibility of the proposed pipeline by testing it on newly recruited nine subjects. Our results show that the pipeline correctly predicts the ventilation mode of 8 out of 9 subjects. The study demonstrates the potential of using image series for lung ventilation mode classification, providing a feasible method for patient prescreening and presenting an alternative form of PFT.
Deep Neural Networks (DNNs) generalization is known to be closely related to the flatness of minima, leading to the development of Sharpness-Aware Minimization (SAM) for seeking flatter minima and better generalization. In this paper, we revisit the loss of SAM and propose a more general method, called WSAM, by incorporating sharpness as a regularization term. We prove its generalization bound through the combination of PAC and Bayes-PAC techniques, and evaluate its performance on various public datasets. The results demonstrate that WSAM achieves improved generalization, or is at least highly competitive, compared to the vanilla optimizer, SAM and its variants. The code is available at https://github.com/intelligent-machine-learning/dlrover/tree/master/atorch/atorch/optimizers.
In recent years, the growing demand for medical imaging diagnosis has brought a significant burden to radiologists. The existing Med-VLP methods provide a solution for automated medical image analysis which learns universal representations from large-scale medical images and reports and benefits downstream tasks without requiring fine-grained annotations. However, the existing methods based on joint image-text reconstruction neglect the importance of cross-modal alignment in conjunction with joint reconstruction, resulting in inadequate cross-modal interaction. In this paper, we propose a unified Med-VLP framework based on Multi-task Paired Masking with Alignment (MPMA) to integrate the cross-modal alignment task into the joint image-text reconstruction framework to achieve more comprehensive cross-modal interaction, while a global and local alignment (GLA) module is designed to assist self-supervised paradigm in obtaining semantic representations with rich domain knowledge. To achieve more comprehensive cross-modal fusion, we also propose a Memory-Augmented Cross-Modal Fusion (MA-CMF) module to fully integrate visual features to assist in the process of report reconstruction. Experimental results show that our approach outperforms previous methods over all downstream tasks, including uni-modal, cross-modal and multi-modal tasks.
Accurately predicting the performance of architecture with small sample training is an important but not easy task. How to analysis and train dataset to overcome overfitting is the core problem we should deal with. Meanwhile if there is the mult-task problem, we should also think about if we can take advantage of their correlation and estimate as fast as we can. In this track, Super Network builds a search space based on ViT-Base. The search space contain depth, num-heads, mpl-ratio and embed-dim. What we done firstly are pre-processing the data based on our understanding of this problem which can reduce complexity of problem and probability of over fitting. Then we tried different kind of models and different way to combine them. Finally we choose stacking ensemble models using GP-NAS with cross validation. Our stacking model ranked 1st in CVPR 2022 Track 2 Challenge.
2D and 3D tumor features are widely used in a variety of medical image analysis tasks. However, for chemotherapy response prediction, the effectiveness between different kinds of 2D and 3D features are not comprehensively assessed, especially in ovarian cancer-related applications. This investigation aims to accomplish such a comprehensive evaluation. For this purpose, CT images were collected retrospectively from 188 advanced-stage ovarian cancer patients. All the metastatic tumors that occurred in each patient were segmented and then processed by a set of six filters. Next, three categories of features, namely geometric, density, and texture features, were calculated from both the filtered results and the original segmented tumors, generating a total of 1595 and 1403 features for the 3D and 2D tumors, respectively. In addition to the conventional single-slice 2D and full-volume 3D tumor features, we also computed the incomplete-3D tumor features, which were achieved by sequentially adding one individual CT slice and calculating the corresponding features. Support vector machine (SVM) based prediction models were developed and optimized for each feature set. 5-fold cross-validation was used to assess the performance of each individual model. The results show that the 2D feature-based model achieved an AUC (area under the ROC curve [receiver operating characteristic]) of 0.84+-0.02. When adding more slices, the AUC first increased to reach the maximum and then gradually decreased to 0.86+-0.02. The maximum AUC was yielded when adding two adjacent slices, with a value of 0.91+-0.01. This initial result provides meaningful information for optimizing machine learning-based decision-making support tools in the future.
The cloud is still a popular platform for distributed deep learning (DL) training jobs since resource sharing in the cloud can improve resource utilization and reduce overall costs. However, such sharing also brings multiple challenges for DL training jobs, e.g., high-priority jobs could impact, even interrupt, low-priority jobs. Meanwhile, most existing distributed DL training systems require users to configure the resources (i.e., the number of nodes and resources like CPU and memory allocated to each node) of jobs manually before job submission and can not adjust the job's resources during the runtime. The resource configuration of a job deeply affect this job's performance (e.g., training throughput, resource utilization, and completion rate). However, this usually leads to poor performance of jobs since users fail to provide optimal resource configuration in most cases. \system~is a distributed DL framework can auto-configure a DL job's initial resources and dynamically tune the job's resources to win the better performance. With elastic capability, \system~can effectively adjusts the resources of a job when there are performance issues detected or a job fails because of faults or eviction. Evaluations results show \system~can outperform manual well-tuned resource configurations. Furthermore, in the production Kubernetes cluster of \company, \system~reduces the medium of job completion time by 31\%, and improves the job completion rate by 6\%, CPU utilization by 15\%, and memory utilization by 20\% compared with manual configuration.
Curating a large scale fully-annotated dataset can be both labour-intensive and expertise-demanding, especially for medical images. To alleviate this problem, we propose to utilize solely scribble annotations for weakly supervised segmentation. Existing solutions mainly leverage selective losses computed solely on annotated areas and generate pseudo gold standard segmentation by propagating labels to adjacent areas. However, these methods could suffer from the inaccurate and sometimes unrealistic pseudo segmentation due to the insufficient supervision and incomplete shape features. Different from previous efforts, we first investigate the principle of ''good scribble annotations'', which leads to efficient scribble forms via supervision maximization and randomness simulation. Furthermore, we introduce regularization terms to encode the spatial relationship and shape prior, where a new formulation is developed to estimate the mixture ratios of label classes. These ratios are critical in identifying the unlabeled pixels for each class and correcting erroneous predictions, thus the accurate estimation lays the foundation for the incorporation of spatial prior. Finally, we integrate the efficient scribble supervision with the prior into a unified framework, denoted as ZScribbleSeg, and apply the method to multiple scenarios. Leveraging only scribble annotations, ZScribbleSeg set new state-of-the-arts on four segmentation tasks using ACDC, MSCMRseg, MyoPS and PPSS datasets.
To improve the generalization of 3D human pose estimators, many existing deep learning based models focus on adding different augmentations to training poses. However, data augmentation techniques are limited to the "seen" pose combinations and hard to infer poses with rare "unseen" joint positions. To address this problem, we present CameraPose, a weakly-supervised framework for 3D human pose estimation from a single image, which can not only be applied on 2D-3D pose pairs but also on 2D alone annotations. By adding a camera parameter branch, any in-the-wild 2D annotations can be fed into our pipeline to boost the training diversity and the 3D poses can be implicitly learned by reprojecting back to 2D. Moreover, CameraPose introduces a refinement network module with confidence-guided loss to further improve the quality of noisy 2D keypoints extracted by 2D pose estimators. Experimental results demonstrate that the CameraPose brings in clear improvements on cross-scenario datasets. Notably, it outperforms the baseline method by 3mm on the most challenging dataset 3DPW. In addition, by combining our proposed refinement network module with existing 3D pose estimators, their performance can be improved in cross-scenario evaluation.