LongDocURL: a Comprehensive Multimodal Long Document Benchmark Integrating Understanding, Reasoning, and Locating

Large vision language models (LVLMs) have improved the document understanding capabilities remarkably, enabling the handling of complex document elements, longer contexts, and a wider range of tasks. However, existing document understanding benchmarks have been limited to handling only a small number of pages and fail to provide a comprehensive analysis of layout elements locating. In this paper, we first define three primary task categories: Long Document Understanding, numerical Reasoning, and cross-element Locating, and then propose a comprehensive benchmark, LongDocURL, integrating above three primary tasks and comprising 20 sub-tasks categorized based on different primary tasks and answer evidences. Furthermore, we develop a semi-automated construction pipeline and collect 2,325 high-quality question-answering pairs, covering more than 33,000 pages of documents, significantly outperforming existing benchmarks. Subsequently, we conduct comprehensive evaluation experiments on both open-source and closed-source models across 26 different configurations, revealing critical performance gaps in this field.

Happy: A Debiased Learning Framework for Continual Generalized Category Discovery

Constantly discovering novel concepts is crucial in evolving environments. This paper explores the underexplored task of Continual Generalized Category Discovery (C-GCD), which aims to incrementally discover new classes from unlabeled data while maintaining the ability to recognize previously learned classes. Although several settings are proposed to study the C-GCD task, they have limitations that do not reflect real-world scenarios. We thus study a more practical C-GCD setting, which includes more new classes to be discovered over a longer period, without storing samples of past classes. In C-GCD, the model is initially trained on labeled data of known classes, followed by multiple incremental stages where the model is fed with unlabeled data containing both old and new classes. The core challenge involves two conflicting objectives: discover new classes and prevent forgetting old ones. We delve into the conflicts and identify that models are susceptible to prediction bias and hardness bias. To address these issues, we introduce a debiased learning framework, namely Happy, characterized by Hardness-aware prototype sampling and soft entropy regularization. For the prediction bias, we first introduce clustering-guided initialization to provide robust features. In addition, we propose soft entropy regularization to assign appropriate probabilities to new classes, which can significantly enhance the clustering performance of new classes. For the harness bias, we present the hardness-aware prototype sampling, which can effectively reduce the forgetting issue for previously seen classes, especially for difficult classes. Experimental results demonstrate our method proficiently manages the conflicts of C-GCD and achieves remarkable performance across various datasets, e.g., 7.5% overall gains on ImageNet-100. Our code is publicly available at https://github.com/mashijie1028/Happy-CGCD.

ModalPrompt:Dual-Modality Guided Prompt for Continual Learning of Large Multimodal Models

Large Multimodal Models (LMMs) exhibit remarkable multi-tasking ability by learning mixed datasets jointly. However, novel tasks would be encountered sequentially in dynamic world, and continually fine-tuning LMMs often leads to performance degrades. To handle the challenges of catastrophic forgetting, existing methods leverage data replay or model expansion, both of which are not specially developed for LMMs and have their inherent limitations. In this paper, we propose a novel dual-modality guided prompt learning framework (ModalPrompt) tailored for multimodal continual learning to effectively learn new tasks while alleviating forgetting of previous knowledge. Concretely, we learn prototype prompts for each task and exploit efficient prompt selection for task identifiers and prompt fusion for knowledge transfer based on image-text supervision. Extensive experiments demonstrate the superiority of our approach, e.g., ModalPrompt achieves +20% performance gain on LMMs continual learning benchmarks with $\times$ 1.42 inference speed refraining from growing training cost in proportion to the number of tasks. The code will be made publically available.

Recoverable Compression: A Multimodal Vision Token Recovery Mechanism Guided by Text Information

With the advancement of large-scale language modeling techniques, large multimodal models combining visual encoders with large language models have demonstrated exceptional performance in various visual tasks. Most of the current large-scale multimodal models achieve this by mapping visual features obtained from the visual encoder into a large language model and using them as inputs alongside text for downstream tasks. Therefore, the number of visual tokens directly affects the training and inference speed of the model. There has been significant work on token pruning for visual transformers, but for large multimodal models, only relying on visual information for token pruning or compression may lead to significant loss of important information. On the other hand, the textual input in the form of a question may contain valuable information that can aid in answering the question, providing additional knowledge to the model. To address the potential oversimplification and excessive pruning that can occur with most purely visual token pruning methods, we propose a text information-guided dynamic visual token recovery mechanism that does not require training. This mechanism leverages the similarity between the question text and visual tokens to recover visually meaningful tokens with important text information while merging other less important tokens. Experimental results demonstrate that our proposed method achieves comparable performance to the original approach while compressing the visual tokens to an average of 10% of the original quantity. Our source code will be made publicly available following acceptance.

Adapting Vision-Language Models to Open Classes via Test-Time Prompt Tuning

Adapting pre-trained models to open classes is a challenging problem in machine learning. Vision-language models fully explore the knowledge of text modality, demonstrating strong zero-shot recognition performance, which is naturally suited for various open-set problems. More recently, some research focuses on fine-tuning such models to downstream tasks. Prompt tuning methods achieved huge improvements by learning context vectors on few-shot data. However, through the evaluation under open-set adaptation setting with the test data including new classes, we find that there exists a dilemma that learned prompts have worse generalization abilities than hand-crafted prompts. In this paper, we consider combining the advantages of both and come up with a test-time prompt tuning approach, which leverages the maximum concept matching (MCM) scores as dynamic weights to generate an input-conditioned prompt for each image during test. Through extensive experiments on 11 different datasets, we show that our proposed method outperforms all comparison methods on average considering both base and new classes. The code is available at https://github.com/gaozhengqing/TTPT

StylePrompter: Enhancing Domain Generalization with Test-Time Style Priors

In real-world applications, the sample distribution at the inference stage often differs from the one at the training stage, causing performance degradation of trained deep models. The research on domain generalization (DG) aims to develop robust algorithms that can improve the generalized performance in unseen domains by training on a few domains. However, the domain-agnostic vision model, trained on a limited number of domains using traditional domain generalization methods, cannot guarantee its effectiveness in dealing with unseen domains. The introduction of language can break the closed cognition space of the vision model, providing additional semantic information that cannot be inferred from vision-only datasets. In this paper, we propose to overcome the challenge in previous DG methods by introducing the style prompt in the language modality to adapt the trained model dynamically. In particular, we train a style prompter to extract style information of the current image into an embedding in the token embedding space and place it in front of the candidate category words as prior knowledge to prompt the model. Our open space partition of the style token embedding space and the hand-crafted style regularization enable the trained style prompter to handle data from unknown domains effectively. Extensive experiments verify the effectiveness of our method and demonstrate state-of-the-art performances on multiple public datasets. Codes will be available after the acceptance of this paper.

PASS++: A Dual Bias Reduction Framework for Non-Exemplar Class-Incremental Learning

Class-incremental learning (CIL) aims to recognize new classes incrementally while maintaining the discriminability of old classes. Most existing CIL methods are exemplar-based, i.e., storing a part of old data for retraining. Without relearning old data, those methods suffer from catastrophic forgetting. In this paper, we figure out two inherent problems in CIL, i.e., representation bias and classifier bias, that cause catastrophic forgetting of old knowledge. To address these two biases, we present a simple and novel dual bias reduction framework that employs self-supervised transformation (SST) in input space and prototype augmentation (protoAug) in deep feature space. On the one hand, SST alleviates the representation bias by learning generic and diverse representations that can transfer across different tasks. On the other hand, protoAug overcomes the classifier bias by explicitly or implicitly augmenting prototypes of old classes in the deep feature space, which poses tighter constraints to maintain previously learned decision boundaries. We further propose hardness-aware prototype augmentation and multi-view ensemble strategies, leading to significant improvements. The proposed framework can be easily integrated with pre-trained models. Without storing any samples of old classes, our method can perform comparably with state-of-the-art exemplar-based approaches which store plenty of old data. We hope to draw the attention of researchers back to non-exemplar CIL by rethinking the necessity of storing old samples in CIL.

WPS-SAM: Towards Weakly-Supervised Part Segmentation with Foundation Models

Segmenting and recognizing diverse object parts is crucial in computer vision and robotics. Despite significant progress in object segmentation, part-level segmentation remains underexplored due to complex boundaries and scarce annotated data. To address this, we propose a novel Weakly-supervised Part Segmentation (WPS) setting and an approach called WPS-SAM, built on the large-scale pre-trained vision foundation model, Segment Anything Model (SAM). WPS-SAM is an end-to-end framework designed to extract prompt tokens directly from images and perform pixel-level segmentation of part regions. During its training phase, it only uses weakly supervised labels in the form of bounding boxes or points. Extensive experiments demonstrate that, through exploiting the rich knowledge embedded in pre-trained foundation models, WPS-SAM outperforms other segmentation models trained with pixel-level strong annotations. Specifically, WPS-SAM achieves 68.93% mIOU and 79.53% mACC on the PartImageNet dataset, surpassing state-of-the-art fully supervised methods by approximately 4% in terms of mIOU.

Fuse, Reason and Verify: Geometry Problem Solving with Parsed Clauses from Diagram

Geometry problem solving (GPS) requires capacities of multi-modal understanding, multi-hop reasoning and theorem knowledge application. In this paper, we propose a neural-symbolic model for plane geometry problem solving (PGPS), named PGPSNet-v2, with three key steps: modal fusion, reasoning process and knowledge verification. In modal fusion, we leverage textual clauses to express fine-grained structural and semantic content of geometry diagram, and fuse diagram with textual problem efficiently through structural-semantic pre-training. For reasoning, we design an explicable solution program to describe the geometric reasoning process, and employ a self-limited decoder to generate solution program autoregressively. To reduce solution errors, a multi-level theorem verifier is proposed to eliminate solutions that do not match geometric principles, alleviating the hallucination of the neural model. We also construct a large-scale geometry problem dataset called PGPS9K, containing fine-grained annotations of textual clauses, solution program and involved knowledge tuples. Extensive experiments on datasets Geometry3K and PGPS9K show that our PGPSNet solver outperforms existing symbolic and neural solvers in GPS performance, while maintaining good explainability and reliability, and the solver components (fusion, reasoning, verification) are all justified effective.

MSPE: Multi-Scale Patch Embedding Prompts Vision Transformers to Any Resolution

Although Vision Transformers (ViTs) have recently advanced computer vision tasks significantly, an important real-world problem was overlooked: adapting to variable input resolutions. Typically, images are resized to a fixed resolution, such as 224x224, for efficiency during training and inference. However, uniform input size conflicts with real-world scenarios where images naturally vary in resolution. Modifying the preset resolution of a model may severely degrade the performance. In this work, we propose to enhance the model adaptability to resolution variation by optimizing the patch embedding. The proposed method, called Multi-Scale Patch Embedding (MSPE), substitutes the standard patch embedding with multiple variable-sized patch kernels and selects the best parameters for different resolutions, eliminating the need to resize the original image. Our method does not require high-cost training or modifications to other parts, making it easy to apply to most ViT models. Experiments in image classification, segmentation, and detection tasks demonstrate the effectiveness of MSPE, yielding superior performance on low-resolution inputs and performing comparably on high-resolution inputs with existing methods.