Zihuan Qiu

Shaoxu Cheng, Kanglei Geng, Chiyuan He et al.

Continual Learning (CL) aims to enable Deep Neural Networks (DNNs) to learn new data without forgetting previously learned knowledge. The key to achieving this goal is to avoid confusion at the feature level, i.e., avoiding confusion within old tasks and between new and old tasks. Previous prototype-based CL methods generate pseudo features for old knowledge replay by adding Gaussian noise to the centroids of old classes. However, the distribution in the feature space exhibits anisotropy during the incremental process, which prevents the pseudo features from faithfully reproducing the distribution of old knowledge in the feature space, leading to confusion in classification boundaries within old tasks. To address this issue, we propose the Distribution-Level Memory Recall (DMR) method, which uses a Gaussian mixture model to precisely fit the feature distribution of old knowledge at the distribution level and generate pseudo features in the next stage. Furthermore, resistance to confusion at the distribution level is also crucial for multimodal learning, as the problem of multimodal imbalance results in significant differences in feature responses between different modalities, exacerbating confusion within old tasks in prototype-based CL methods. Therefore, we mitigate the multi-modal imbalance problem by using the Inter-modal Guidance and Intra-modal Mining (IGIM) method to guide weaker modalities with prior information from dominant modalities and further explore useful information within modalities. For the second key, We propose the Confusion Index to quantitatively describe a model's ability to distinguish between new and old tasks, and we use the Incremental Mixup Feature Enhancement (IMFE) method to enhance pseudo features with new sample features, alleviating classification confusion between new and old knowledge.

Chiyuan He, Zihuan Qiu, Fanman Meng et al.

Continual learning of pretrained vision-language models (VLMs) is prone to catastrophic forgetting, yet current approaches adapt to new tasks without explicitly preserving the cross-modal semantic geometry inherited from pretraining and previous stages, allowing new-task supervision to induce geometric distortion. We observe that the most pronounced drift tends to concentrate in vulnerable neighborhoods near the old-new semantic interface, where shared visual patterns are easily re-explained by new textual semantics. To address this under an exemplar-free constraint, we propose Semantic Geometry Preservation for Continual Learning (SeGP-CL). SeGP-CL first probes the drift-prone region by constructing a compact set of adversarial anchors with dual-targeted projected gradient descent (DPGD), which drives selected new-task seeds toward old-class semantics while remaining faithful in raw visual space. During training, we preserve cross-modal structure by anchor-guided cross-modal geometry distillation (ACGD), and stabilize the textual reference frame across tasks via a lightweight text semantic-geometry regularization (TSGR). After training, we estimate anchor-induced raw-space drift to transfer old visual prototypes and perform dual-path inference by fusing cross-modal and visual cues. Extensive experiments on five continual learning benchmarks demonstrate that SeGP-CL consistently improves stability and forward transfer, achieving state-of-the-art performance while better preserving semantic geometry of VLMs.

Zihuan Qiu, Yi Xu, Chiyuan He et al.

Continual model merging integrates independently fine-tuned models sequentially without access to the original training data, offering a scalable and efficient solution for continual learning. However, existing methods face two critical challenges: parameter interference among tasks, which leads to catastrophic forgetting, and limited adaptability to evolving test distributions. To address these issues, we introduce the task of Test-Time Continual Model Merging (TTCMM), which leverages a small set of unlabeled test samples during inference to alleviate parameter conflicts and handle distribution shifts. We propose MINGLE, a novel framework for TTCMM. MINGLE employs a mixture-of-experts architecture with parameter-efficient, low-rank experts, which enhances adaptability to evolving test distributions while dynamically merging models to mitigate conflicts. To further reduce forgetting, we propose Null-Space Constrained Gating, which restricts gating updates to subspaces orthogonal to prior task representations, thereby suppressing activations on old tasks and preserving past knowledge. We further introduce an Adaptive Relaxation Strategy that adjusts constraint strength dynamically based on interference signals observed during test-time adaptation, striking a balance between stability and adaptability. Extensive experiments on standard continual merging benchmarks demonstrate that MINGLE achieves robust generalization, significantly reduces forgetting, and consistently surpasses previous state-of-the-art methods by 7-9% on average across diverse task orders. Our code is available at: https://github.com/zihuanqiu/MINGLE

Qingyu Xu, Runtong Zhang, Zihuan Qiu et al.

Object detection in fire rescue scenarios is importance for command and decision-making in firefighting operations. However, existing research still suffers from two main limitations. First, current work predominantly focuses on environments such as mountainous or forest areas, while paying insufficient attention to urban rescue scenes, which are more frequent and structurally complex. Second, existing detection systems include a limited number of classes, such as flames and smoke, and lack a comprehensive system covering key targets crucial for command decisions, such as fire trucks and firefighters. To address the above issues, this paper first constructs a new dataset named "FireRescue" for rescue command, which covers multiple rescue scenarios, including urban, mountainous, forest, and water areas, and contains eight key categories such as fire trucks and firefighters, with a total of 15,980 images and 32,000 bounding boxes. Secondly, to tackle the problems of inter-class confusion and missed detection of small targets caused by chaotic scenes, diverse targets, and long-distance shooting, this paper proposes an improved model named FRS-YOLO. On the one hand, the model introduces a plug-and-play multidi-mensional collaborative enhancement attention module, which enhances the discriminative representation of easily confused categories (e.g., fire trucks vs. ordinary trucks) through cross-dimensional feature interaction. On the other hand, it integrates a dynamic feature sampler to strengthen high-response foreground features, thereby mitigating the effects of smoke occlusion and background interference. Experimental results demonstrate that object detection in fire rescue scenarios is highly challenging, and the proposed method effectively improves the detection performance of YOLO series models in this context.

Zihuan Qiu, Zhichuan Wang, Miaomiao Zhang et al.

Colorectal cancer (CRC) is one of the most common fatal cancer in the world. Polypectomy can effectively interrupt the progression of adenoma to adenocarcinoma, thus reducing the risk of CRC development. Colonoscopy is the primary method to find colonic polyps. However, due to the different sizes of polyps and the unclear boundary between polyps and their surrounding mucosa, it is challenging to segment polyps accurately. To address this problem, we design a Boundary Distribution Guided Network (BDG-Net) for accurate polyp segmentation. Specifically, under the supervision of the ideal Boundary Distribution Map (BDM), we use Boundary Distribution Generate Module (BDGM) to aggregate high-level features and generate BDM. Then, BDM is sent to the Boundary Distribution Guided Decoder (BDGD) as complementary spatial information to guide the polyp segmentation. Moreover, a multi-scale feature interaction strategy is adopted in BDGD to improve the segmentation accuracy of polyps with different sizes. Extensive quantitative and qualitative evaluations demonstrate the effectiveness of our model, which outperforms state-of-the-art models remarkably on five public polyp datasets while maintaining low computational complexity. Code: https://github.com/zihuanqiu/BDG-Net

Zihuan Qiu, Yi Xu, Fanman Meng et al.

Class Incremental Learning (CIL) aims to sequentially acquire knowledge of new classes without forgetting previously learned ones. Despite recent progress, current CIL methods still exhibit significant performance gaps compared to their oracle counterparts-models trained with full access to historical data. Inspired by recent insights on Linear Mode Connectivity (LMC), we revisit the geometric properties of oracle solutions in CIL and uncover a fundamental observation: these oracle solutions typically maintain low-loss linear connections to the optimum of previous tasks. Motivated by this finding, we propose Increment Vector Transformation (IVT), a novel plug-and-play framework designed to mitigate catastrophic forgetting during training. Rather than directly following CIL updates, IVT periodically teleports the model parameters to transformed solutions that preserve linear connectivity to previous task optimum. By maintaining low-loss along these connecting paths, IVT effectively ensures stable performance on previously learned tasks. The transformation is efficiently approximated using diagonal Fisher Information Matrices, making IVT suitable for both exemplar-free and exemplar-based scenarios, and compatible with various initialization strategies. Extensive experiments on CIFAR-100, FGVCAircraft, ImageNet-Subset, and ImageNet-Full demonstrate that IVT consistently enhances the performance of strong CIL baselines. Specifically, on CIFAR-100, IVT improves the last accuracy of the PASS baseline by +5.12% and reduces forgetting by 2.54%. For the CLIP-pre-trained SLCA baseline on FGVCAircraft, IVT yields gains of +14.93% in average accuracy and +21.95% in last accuracy. The code will be released.

Zihuan Qiu, Lei Wang, Yang Cao et al.

Data-free continual model merging (DFCMM) aims to fuse independently fine-tuned models into a single backbone that evolves with incoming tasks without accessing task data. This paper formulate two fundamental desiderata for DFCMM: transparency, avoiding interference with earlier tasks, and fidelity, adapting faithfully to each new task. This poses a challenge that existing approaches fail to address: how to bridge data-level desiderata with parameter-space optimization to ensure transparency and fidelity in the absence of task data. To this end, we propose NUFILT (NUll-space FILTering), a data-free framework that directly links these desiderata to optimization. Our key observation is that task vectors approximately align with representation subspaces, providing structural surrogates for enforcing transparency and fidelity. Accordingly, we design a null-space projector that preserves prior responses by filtering out overlapping components of new task vectors, thereby ensuring transparency, and a lightweight LoRA adapter that injects complementary task-specific signals, enabling fidelity in adapting to new tasks. The adapter is trained with a projection-based surrogate loss to retain consistency with previous knowledge while introducing novel directions. This joint filtering-adaptation process allows the backbone to absorb new knowledge while retaining existing behaviors, and the updates are finally fused back in a layer-wise linear fashion without extra parameters or inference cost. Theoretically, we establish approximate subspace alignment guarantees that justify null-space filtering. Empirically, NUFILT achieves state-of-the-art performance with minimal forgetting on both vision and NLP benchmarks, improving average accuracy by 4-7% over OPCM and WUDI-Merging, while narrowing the gap to fine-tuning and reducing computation overhead.

Yukun Chen, Zihuan Qiu, Fanman Meng et al.

Unlike traditional Multimodal Class-Incremental Learning (MCIL) methods that focus only on vision and text, this paper explores MCIL across vision, audio and text modalities, addressing challenges in integrating complementary information and mitigating catastrophic forgetting. To tackle these issues, we propose an MCIL method based on multimodal pre-trained models. Firstly, a Multimodal Incremental Feature Extractor (MIFE) based on Mixture-of-Experts (MoE) structure is introduced to achieve effective incremental fine-tuning for AudioCLIP. Secondly, to enhance feature discriminability and generalization, we propose an Adaptive Audio-Visual Fusion Module (AAVFM) that includes a masking threshold mechanism and a dynamic feature fusion mechanism, along with a strategy to enhance text diversity. Thirdly, a novel multimodal class-incremental contrastive training loss is proposed to optimize cross-modal alignment in MCIL. Finally, two MCIL-specific evaluation metrics are introduced for comprehensive assessment. Extensive experiments on three multimodal datasets validate the effectiveness of our method.

Chiyuan He, Zihuan Qiu, Fanman Meng et al.

Continual learning of vision-language models (VLMs) focuses on leveraging cross-modal pretrained knowledge to incrementally adapt to expanding downstream tasks and datasets, while tackling the challenge of knowledge forgetting. Existing research often focuses on connecting visual features with specific class text in downstream tasks, overlooking the latent relationships between general and specialized knowledge. Our findings reveal that forcing models to optimize inappropriate visual-text matches exacerbates forgetting of VLM's recognition ability. To tackle this issue, we propose DesCLIP, which leverages general attribute (GA) descriptions to guide the understanding of specific class objects, enabling VLMs to establish robust vision-GA-class trilateral associations rather than relying solely on vision-class connections. Specifically, we introduce a language assistant to generate concrete GA description candidates via proper request prompts. Then, an anchor-based embedding filter is designed to obtain highly relevant GA description embeddings, which are leveraged as the paired text embeddings for visual-textual instance matching, thereby tuning the visual encoder. Correspondingly, the class text embeddings are gradually calibrated to align with these shared GA description embeddings. Extensive experiments demonstrate the advancements and efficacy of our proposed method, with comprehensive empirical evaluations highlighting its superior performance in VLM-based recognition compared to existing continual learning methods.