Lifelong learning on Whole Slide Images (WSIs) aims to train or fine-tune a unified model sequentially on cancer-related tasks, reducing the resources and effort required for data transfer and processing, especially given the gigabyte-scale size of WSIs. In this paper, we introduce MergeSlide, a simple yet effective framework that treats lifelong learning as a model-merging problem by leveraging a vision–language pathology foundation model. When a new task arrives, it is ❶ defined with class-aware prompts, ❷ fine-tuned for a few epochs in a classifier-free manner, and ❸ merged into a unified model using an orthogonal continual-merging strategy that preserves performance and mitigates catastrophic forgetting. For inference under the class-incremental learning (CLASS-IL) setting, where task identity is unknown, we introduce Task-to-Class Prompt-aligned (TCP) inference. Specifically, TCP first identifies the most relevant task using task-level prompts and then applies the corresponding class-aware prompts to generate predictions. To evaluate MergeSlide, we conduct experiments on a stream of six TCGA datasets. The results show that MergeSlide outperforms both rehearsal-based continual learning and vision-language zero-shot baselines. Code and data are available at https://github.com/caodoanh2001/MergeSlide.

The enhanced representational power and broad applicability of deep learning models have attracted significant interest from the research community in recent years. However, these models often struggle to perform effectively under domain shift conditions, where the training data (the source domain) is related to but exhibits different distributions from the testing data (the target domain). To address this challenge, previous studies have attempted to reduce the domain gap between source and target data by incorporating a few labeled target samples during training—a technique known as semi-supervised domain adaptation (SSDA). While this strategy has demonstrated notable improvements in classification performance, the network architectures used in these approaches primarily focus on exploiting the features of individual images, leaving room for improvement in capturing rich representations. In this study, we introduce a Hierarchical Graph of Nodes designed to simultaneously present representations at both feature and category levels. At the feature level, we introduce a local graph to identify the most relevant patches within an image, facilitating adaptability to defined main object representations. At the category level, we employ a global graph to aggregate the features from samples within the same category, thereby enriching overall representations. Extensive experiments on widely used SSDA benchmark datasets, including Office-Home, DomainNet, and VisDA2017, demonstrate that both quantitative and qualitative results substantiate the effectiveness of HiGDA, establishing it as a new state-of-the-art method.

Most domain adaptation (DA) methods are based on either a convolutional neural networks (CNNs) or a vision transformers (ViTs). They align the distribution differences between domains as encoders without considering their unique characteristics. For instance, ViT excels in accuracy due to its superior ability to capture global representations, while CNN has an advantage in capturing local representations. This fact has led us to design a hybrid method to fully take advantage of both ViT and CNN, called Explicitly Class-specific Boundaries (ECB). ECB learns CNN on ViT to combine their distinct strengths. In particular, we leverage ViT's properties to explicitly find class-specific decision boundaries by maximizing the discrepancy between the out-puts of the two classifiers to detect target samples far from the source support. In contrast, the CNN encoder clusters target features based on the previously defined class-specific boundaries by minimizing the discrepancy between the prob-abilities of the two classifiers. Finally, ViT and CNN mutually exchange knowledge to improve the quality of pseudo labels and reduce the knowledge discrepancies of these models. Compared to conventional DA methods, our ECB achieves superior performance, which verifies its effectiveness in this hybrid model. The project website can be found here.

The motivation of the semi-supervised domain adaptation (SSDA) is to train a model by leveraging knowledge acquired from the plentiful labeled source combined with extremely scarce labeled target data to achieve the lowest error on the unlabeled target data at the testing time. However, due to inter-domain and intra-domain discrepancies, the improvement of classification accuracy is limited. To solve these, we propose the Trico-training method that utilizes a multilayer perceptron (MLP) classifier and two graph convolutional network (GCN) classifiers called interview GCN and intra-view GCN classifiers. The first co-training strategy exploits a correlation between MLP and inter-view GCN classifiers to minimize the inter-domain discrepancy, in which the inter-view GCN classifier provides its pseudo labels to teach the MLP classifier, which encourages class representation alignment across domains. In contrast, the MLP classifier gives feedback to the inter-view GCN classifier by using a new concept, ‘pseudo-edge’, for neighbor’s feature aggregation. Doing this increases the data structure mining ability of the inter-view GCN classifier; thus, the quality of generated pseudo labels is improved. The second co-training strategy between MLP and intra-view GCN is conducted in a similar way to reduce the intra-domain discrepancy by enhancing the correlation between labeled and unlabeled target data. Due to an imbalance in classification accuracy between inter-view and intra-view GCN classifiers, we propose the third co-training strategy that encourages them to cooperate to address this problem. We verify the effectiveness of the proposed method on three standard SSDA benchmark datasets: Office-31, Office-Home, and DomainNet. The extended experimental results show that our method surpasses the prior state-of-the-art approaches in SSDA.

Multitarget domain adaptation (MTDA) is a transfer learning task that uses knowledge extracted from a labeled source domain to adapt across multiple unlabeled target domains. The MTDA setting is more complicated than the single-source-single-target domain adaptation (S3TDA) setting because domain shift not only exists in each pair of a source–target domain but also exists among different target domains. In addition, multiple-target domains have their own unique characteristics because they are often collected from various conditions. The semantic information in each target domain can be damaged when they are naïvely merged into a single-target domain. Therefore, the trained model struggles to distinguish between representations in the combined target domain, which degrades the classification performance. Furthermore, the knowledge transferability from the source domain to multiple-target domains in prior studies leaves room for improvement because they only focus on exploiting the relationship of source–target pairs while failing to consider the correlation among multiple-target domains. This article introduces an easy-to-hard adaption structure to solve these problems in MTDA. The proposed method consists of three components: Extracting source representations, Hierarchical intratarget feature Alignment, and Collaborative intertarget feature Alignment, called EHACA. These components are used to encode the semantic information in each target domain and explore the relationships between the source and target domains, and among different target domains. The proposed method shows outstanding classification performance over five remote sensing datasets of MTDA tasks, surpassing state-of-the-art approaches in most experimental scenarios.

In this work, we observed that a nonlinear line could be expressed with a set of linear lines. We propose a novel lane detection method with polyline-based reconstruction based on this hypothesis. We define the optimal principal points with a new metric, the principal score, to generate the polyline. According to the principal score, we select principal points having a high influence on lane reconstruction and simply reproduce the target lane by connecting them. Additionally, conventional methods predict a fixed number of parameters to express each lane. However, this can limit an ability to represent a lane curvature and cause inaccurate detection results. Therefore, we set the number of principal points to be dynamically changed depending on the lane curvature to solve this problem. This allows the model to make flexible detection results reflecting the characteristics of each lane. We also propose a training strategy with a new piece-wise linear equation-based loss function. With this strategy, the model is fine-tuned to predict the principal points representing the curved parts of the lane well. Last, we propose a spatial context-aware feature flip fusion module to exploit the symmetric property of road images. This module helps the model selectively utilize the spatial context in the flipped feature map based on the lane density. We effectively reduce the adverse effects, especially the false positives of the existing feature flip fusion module misaligned on asymmetrical images. The experiments show that the proposed method provides competitive lane detection results compared to state-of-the-art methods.

Due to the unique characteristics of remote sensing (RS) data, it is challenging to collect richer labeled samples for training the deep learning model compared with the natural image data. To solve this problem, recently, multi-source-single-target (MS2T) scenarios have started receiving significant attention in which the knowledge from multiple sources is integrated to transfer to a target domain with the assumption that label spaces of each source and target domain are the same. However, in real-world applications, it can be challenging to find a source domain that completely includes all classes of the target domain. Therefore, to cover all class information of the target domain, they often naïvely merge multiple sources into a complete single source. However, each source domain typically has a different data distribution; thus, the semantic information of each source domain can be damaged, leading to degrading the classification accuracy of the target domain. To address this problem, we propose a unified framework termed Multi-Expert Collaboration for Knowledge Adaptation (MECKA) from various sources. MECKA includes two main processes: multiple-view generation and collaborative learning. Multiview learning plays an essential role in preserving the unique characteristics of each source domain. In contrast, collaborative learning is responsible for connecting these views that leverage complementary information from each other to perform on an unseen target domain robustly. Experimental results showed that the proposed method achieved the best classification accuracy on remote sensing scene benchmark datasets on both complete and incomplete multisource unsupervised domain adaptation (UDA) tasks compared to benchmark methods.