Daily Papers

We propose a novel architecture for labelling character sequences that achieves state-of-the-art results on the Tashkeela Arabic diacritization benchmark. The core is a two-level recurrence hierarchy that operates on the word and character levels separately---enabling faster training and inference than comparable traditional models. A cross-level attention module further connects the two, and opens the door for network interpretability. The task module is a softmax classifier that enumerates valid combinations of diacritics. This architecture can be extended with a recurrent decoder that optionally accepts priors from partially diacritized text, which improves results. We employ extra tricks such as sentence dropout and majority voting to further boost the final result. Our best model achieves a WER of 5.34%, outperforming the previous state-of-the-art with a 30.56% relative error reduction.

Scene text editing, a subfield of image editing, requires modifying texts in images while preserving style consistency and visual coherence with the surrounding environment. While diffusion-based methods have shown promise in text generation, they still struggle to produce high-quality results. These methods often generate distorted or unrecognizable characters, particularly when dealing with complex characters like Chinese. In such systems, characters are composed of intricate stroke patterns and spatial relationships that must be precisely maintained. We present GlyphMastero, a specialized glyph encoder designed to guide the latent diffusion model for generating texts with stroke-level precision. Our key insight is that existing methods, despite using pretrained OCR models for feature extraction, fail to capture the hierarchical nature of text structures - from individual strokes to stroke-level interactions to overall character-level structure. To address this, our glyph encoder explicitly models and captures the cross-level interactions between local-level individual characters and global-level text lines through our novel glyph attention module. Meanwhile, our model implements a feature pyramid network to fuse the multi-scale OCR backbone features at the global-level. Through these cross-level and multi-scale fusions, we obtain more detailed glyph-aware guidance, enabling precise control over the scene text generation process. Our method achieves an 18.02\% improvement in sentence accuracy over the state-of-the-art multi-lingual scene text editing baseline, while simultaneously reducing the text-region Fr\'echet inception distance by 53.28\%.

While transformers have greatly boosted performance in semantic segmentation, domain adaptive transformers are not yet well explored. We identify that the domain gap can cause discrepancies in self-attention. Due to this gap, the transformer attends to spurious regions or pixels, which deteriorates accuracy on the target domain. We propose to perform adaptation on attention maps with cross-domain attention layers that share features between the source and the target domains. Specifically, we impose consistency between predictions from cross-domain attention and self-attention modules to encourage similar distribution in the attention and output of the model across domains, i.e., attention-level and output-level alignment. We also enforce consistency in attention maps between different augmented views to further strengthen the attention-based alignment. Combining these two components, our method mitigates the discrepancy in attention maps across domains and further boosts the performance of the transformer under unsupervised domain adaptation settings. Our model outperforms the existing state-of-the-art baseline model on three widely used benchmarks, including GTAV-to-Cityscapes by 1.3 percent point (pp), Synthia-to-Cityscapes by 0.6 pp, and Cityscapes-to-ACDC by 1.1 pp, on average. Additionally, we verify the effectiveness and generalizability of our method through extensive experiments. Our code will be publicly available.

This paper provides a novel 3D medical image segmentation model structure called nnY-Net. This name comes from the fact that our model adds a cross-attention module at the bottom of the U-net structure to form a Y structure. We integrate the advantages of the two latest SOTA models, MedNeXt and SwinUNETR, and use Swin Transformer as the encoder and ConvNeXt as the decoder to innovatively design the Swin-NeXt structure. Our model uses the lowest-level feature map of the encoder as Key and Value and uses patient features such as pathology and treatment information as Query to calculate the attention weights in a Cross Attention module. Moreover, we simplify some pre- and post-processing as well as data enhancement methods in 3D image segmentation based on the dynUnet and nnU-net frameworks. We integrate our proposed Swin-NeXt with Cross-Attention framework into this framework. Last, we construct a DiceFocalCELoss to improve the training efficiency for the uneven data convergence of voxel classification.

Medical image segmentation plays a crucial role in assisting healthcare professionals with accurate diagnoses and enabling automated diagnostic processes. Traditional convolutional neural networks (CNNs) often struggle with capturing long-range dependencies, while transformer-based architectures, despite their effectiveness, come with increased computational complexity. Recent efforts have focused on combining CNNs and transformers to balance performance and efficiency, but existing approaches still face challenges in achieving high segmentation accuracy while maintaining low computational costs. Furthermore, many methods underutilize the CNN encoder's capability to capture local spatial information, concentrating primarily on mitigating long-range dependency issues. To address these limitations, we propose QTSeg, a novel architecture for medical image segmentation that effectively integrates local and global information. QTSeg features a dual-mix attention decoder designed to enhance segmentation performance through: (1) a cross-attention mechanism for improved feature alignment, (2) a spatial attention module to capture long-range dependencies, and (3) a channel attention block to learn inter-channel relationships. Additionally, we introduce a multi-level feature distribution module, which adaptively balances feature propagation between the encoder and decoder, further boosting performance. Extensive experiments on five publicly available datasets covering diverse segmentation tasks, including lesion, polyp, breast cancer, cell, and retinal vessel segmentation, demonstrate that QTSeg outperforms state-of-the-art methods across multiple evaluation metrics while maintaining lower computational costs. Our implementation can be found at: https://github.com/tpnam0901/QTSeg (v1.0.0)

In standard hospital blood tests, the traditional process requires doctors to manually isolate leukocytes from microscopic images of patients' blood using microscopes. These isolated leukocytes are then categorized via automatic leukocyte classifiers to determine the proportion and volume of different types of leukocytes present in the blood samples, aiding disease diagnosis. This methodology is not only time-consuming and labor-intensive, but it also has a high propensity for errors due to factors such as image quality and environmental conditions, which could potentially lead to incorrect subsequent classifications and misdiagnosis. To address these issues, this paper proposes an innovative method of leukocyte detection: the Multi-level Feature Fusion and Deformable Self-attention DETR (MFDS-DETR). To tackle the issue of leukocyte scale disparity, we designed the High-level Screening-feature Fusion Pyramid (HS-FPN), enabling multi-level fusion. This model uses high-level features as weights to filter low-level feature information via a channel attention module and then merges the screened information with the high-level features, thus enhancing the model's feature expression capability. Further, we address the issue of leukocyte feature scarcity by incorporating a multi-scale deformable self-attention module in the encoder and using the self-attention and cross-deformable attention mechanisms in the decoder, which aids in the extraction of the global features of the leukocyte feature maps. The effectiveness, superiority, and generalizability of the proposed MFDS-DETR method are confirmed through comparisons with other cutting-edge leukocyte detection models using the private WBCDD, public LISC and BCCD datasets. Our source code and private WBCCD dataset are available at https://github.com/JustlfC03/MFDS-DETR.

Text-to-image based object customization, aiming to generate images with the same identity (ID) as objects of interest in accordance with text prompts and reference images, has made significant progress. However, recent customizing research is dominated by specialized tasks, such as human customization or virtual try-on, leaving a gap in general object customization. To this end, we introduce AnyMaker, an innovative zero-shot object customization framework capable of generating general objects with high ID fidelity and flexible text editability. The efficacy of AnyMaker stems from its novel general ID extraction, dual-level ID injection, and ID-aware decoupling. Specifically, the general ID extraction module extracts sufficient ID information with an ensemble of self-supervised models to tackle the diverse customization tasks for general objects. Then, to provide the diffusion UNet with the extracted ID as much while not damaging the text editability in the generation process, we design a global-local dual-level ID injection module, in which the global-level semantic ID is injected into text descriptions while the local-level ID details are injected directly into the model through newly added cross-attention modules. In addition, we propose an ID-aware decoupling module to disentangle ID-related information from non-ID elements in the extracted representations for high-fidelity generation of both identity and text descriptions. To validate our approach and boost the research of general object customization, we create the first large-scale general ID dataset, Multi-Category ID-Consistent (MC-IDC) dataset, with 315k text-image samples and 10k categories. Experiments show that AnyMaker presents remarkable performance in general object customization and outperforms specialized methods in corresponding tasks. Code and dataset will be released soon.

Realistic image super-resolution (Real-ISR) aims to reproduce perceptually realistic image details from a low-quality input. The commonly used adversarial training based Real-ISR methods often introduce unnatural visual artifacts and fail to generate realistic textures for natural scene images. The recently developed generative stable diffusion models provide a potential solution to Real-ISR with pre-learned strong image priors. However, the existing methods along this line either fail to keep faithful pixel-wise image structures or resort to extra skipped connections to reproduce details, which requires additional training in image space and limits their extension to other related tasks in latent space such as image stylization. In this work, we propose a pixel-aware stable diffusion (PASD) network to achieve robust Real-ISR as well as personalized stylization. In specific, a pixel-aware cross attention module is introduced to enable diffusion models perceiving image local structures in pixel-wise level, while a degradation removal module is used to extract degradation insensitive features to guide the diffusion process together with image high level information. By simply replacing the base diffusion model with a personalized one, our method can generate diverse stylized images without the need to collect pairwise training data. PASD can be easily integrated into existing diffusion models such as Stable Diffusion. Experiments on Real-ISR and personalized stylization demonstrate the effectiveness of our proposed approach. The source code and models can be found at https://github.com/yangxy/PASD.

Text-to-video (T2V) generation has gained significant attention due to its wide applications to video generation, editing, enhancement and translation, \etc. However, high-quality (HQ) video synthesis is extremely challenging because of the diverse and complex motions existed in real world. Most existing works struggle to address this problem by collecting large-scale HQ videos, which are inaccessible to the community. In this work, we show that publicly available limited and low-quality (LQ) data are sufficient to train a HQ video generator without recaptioning or finetuning. We factorize the whole T2V generation process into two steps: generating an image conditioned on a highly descriptive caption, and synthesizing the video conditioned on the generated image and a concise caption of motion details. Specifically, we present Factorized-Dreamer, a factorized spatiotemporal framework with several critical designs for T2V generation, including an adapter to combine text and image embeddings, a pixel-aware cross attention module to capture pixel-level image information, a T5 text encoder to better understand motion description, and a PredictNet to supervise optical flows. We further present a noise schedule, which plays a key role in ensuring the quality and stability of video generation. Our model lowers the requirements in detailed captions and HQ videos, and can be directly trained on limited LQ datasets with noisy and brief captions such as WebVid-10M, largely alleviating the cost to collect large-scale HQ video-text pairs. Extensive experiments in a variety of T2V and image-to-video generation tasks demonstrate the effectiveness of our proposed Factorized-Dreamer. Our source codes are available at https://github.com/yangxy/Factorized-Dreamer/.

Recent text-to-image models produce high-quality results but still struggle with precise visual control, balancing multimodal inputs, and requiring extensive training for complex multimodal image generation. To address these limitations, we propose MENTOR, a novel autoregressive (AR) framework for efficient Multimodal-conditioned Tuning for Autoregressive multimodal image generation. MENTOR combines an AR image generator with a two-stage training paradigm, enabling fine-grained, token-level alignment between multimodal inputs and image outputs without relying on auxiliary adapters or cross-attention modules. The two-stage training consists of: (1) a multimodal alignment stage that establishes robust pixel- and semantic-level alignment, followed by (2) a multimodal instruction tuning stage that balances the integration of multimodal inputs and enhances generation controllability. Despite modest model size, suboptimal base components, and limited training resources, MENTOR achieves strong performance on the DreamBench++ benchmark, outperforming competitive baselines in concept preservation and prompt following. Additionally, our method delivers superior image reconstruction fidelity, broad task adaptability, and improved training efficiency compared to diffusion-based methods. Dataset, code, and models are available at: https://github.com/HaozheZhao/MENTOR

Text-driven video generation witnesses rapid progress. However, merely using text prompts is not enough to depict the desired subject appearance that accurately aligns with users' intents, especially for customized content creation. In this paper, we study the task of video generation with image prompts, which provide more accurate and direct content control beyond the text prompts. Specifically, we propose a feed-forward framework VideoBooth, with two dedicated designs: 1) We propose to embed image prompts in a coarse-to-fine manner. Coarse visual embeddings from image encoder provide high-level encodings of image prompts, while fine visual embeddings from the proposed attention injection module provide multi-scale and detailed encoding of image prompts. These two complementary embeddings can faithfully capture the desired appearance. 2) In the attention injection module at fine level, multi-scale image prompts are fed into different cross-frame attention layers as additional keys and values. This extra spatial information refines the details in the first frame and then it is propagated to the remaining frames, which maintains temporal consistency. Extensive experiments demonstrate that VideoBooth achieves state-of-the-art performance in generating customized high-quality videos with subjects specified in image prompts. Notably, VideoBooth is a generalizable framework where a single model works for a wide range of image prompts with feed-forward pass.

Understanding 3D scenes semantically and spatially is crucial for the safe navigation of robots and autonomous vehicles, aiding obstacle avoidance and accurate trajectory planning. Camera-based 3D semantic occupancy prediction, which infers complete voxel grids from 2D images, is gaining importance in robot vision for its resource efficiency compared to 3D sensors. However, this task inherently suffers from a 2D-3D discrepancy, where objects of the same size in 3D space appear at different scales in a 2D image depending on their distance from the camera due to perspective projection. To tackle this issue, we propose a novel framework called VPOcc that leverages a vanishing point (VP) to mitigate the 2D-3D discrepancy at both the pixel and feature levels. As a pixel-level solution, we introduce a VPZoomer module, which warps images by counteracting the perspective effect using a VP-based homography transformation. In addition, as a feature-level solution, we propose a VP-guided cross-attention (VPCA) module that performs perspective-aware feature aggregation, utilizing 2D image features that are more suitable for 3D space. Lastly, we integrate two feature volumes extracted from the original and warped images to compensate for each other through a spatial volume fusion (SVF) module. By effectively incorporating VP into the network, our framework achieves improvements in both IoU and mIoU metrics on SemanticKITTI and SSCBench-KITTI360 datasets. Additional details are available at https://vision3d-lab.github.io/vpocc/.

The remarkable development of text-to-image generation models has raised notable security concerns, such as the infringement of portrait rights and the generation of inappropriate content. Concept erasure has been proposed to remove the model's knowledge about protected and inappropriate concepts. Although many methods have tried to balance the efficacy (erasing target concepts) and specificity (retaining irrelevant concepts), they can still generate abundant erasure concepts under the steering of semantically related inputs. In this work, we propose RealEra to address this "concept residue" issue. Specifically, we first introduce the mechanism of neighbor-concept mining, digging out the associated concepts by adding random perturbation into the embedding of erasure concept, thus expanding the erasing range and eliminating the generations even through associated concept inputs. Furthermore, to mitigate the negative impact on the generation of irrelevant concepts caused by the expansion of erasure scope, RealEra preserves the specificity through the beyond-concept regularization. This makes irrelevant concepts maintain their corresponding spatial position, thereby preserving their normal generation performance. We also employ the closed-form solution to optimize weights of U-Net for the cross-attention alignment, as well as the prediction noise alignment with the LoRA module. Extensive experiments on multiple benchmarks demonstrate that RealEra outperforms previous concept erasing methods in terms of superior erasing efficacy, specificity, and generality. More details are available on our project page https://realerasing.github.io/RealEra/ .

Video personalization methods allow us to synthesize videos with specific concepts such as people, pets, and places. However, existing methods often focus on limited domains, require time-consuming optimization per subject, or support only a single subject. We present Video Alchemist - a video model with built-in multi-subject, open-set personalization capabilities for both foreground objects and background, eliminating the need for time-consuming test-time optimization. Our model is built on a new Diffusion Transformer module that fuses each conditional reference image and its corresponding subject-level text prompt with cross-attention layers. Developing such a large model presents two main challenges: dataset and evaluation. First, as paired datasets of reference images and videos are extremely hard to collect, we sample selected video frames as reference images and synthesize a clip of the target video. However, while models can easily denoise training videos given reference frames, they fail to generalize to new contexts. To mitigate this issue, we design a new automatic data construction pipeline with extensive image augmentations. Second, evaluating open-set video personalization is a challenge in itself. To address this, we introduce a personalization benchmark that focuses on accurate subject fidelity and supports diverse personalization scenarios. Finally, our extensive experiments show that our method significantly outperforms existing personalization methods in both quantitative and qualitative evaluations.

We present a novel text-to-speech (TTS) system, namely SupertonicTTS, for improved scalability and efficiency in speech synthesis. SupertonicTTS is comprised of three components: a speech autoencoder for continuous latent representation, a text-to-latent module leveraging flow-matching for text-to-latent mapping, and an utterance-level duration predictor. To enable a lightweight architecture, we employ a low-dimensional latent space, temporal compression of latents, and ConvNeXt blocks. We further simplify the TTS pipeline by operating directly on raw character-level text and employing cross-attention for text-speech alignment, thus eliminating the need for grapheme-to-phoneme (G2P) modules and external aligners. In addition, we introduce context-sharing batch expansion that accelerates loss convergence and stabilizes text-speech alignment. Experimental results demonstrate that SupertonicTTS achieves competitive performance while significantly reducing architectural complexity and computational overhead compared to contemporary TTS models. Audio samples demonstrating the capabilities of SupertonicTTS are available at: https://supertonictts.github.io/.

In multimodal large language models (MLLMs), vision transformers (ViTs) are widely employed for visual encoding. However, their performance in solving universal MLLM tasks is not satisfactory. We attribute it to a lack of information from diverse visual levels, impeding alignment with the various semantic granularity required for language generation. To address this issue, we present LLaVA-UHD v2, an advanced MLLM centered around a Hierarchical window transformer that enables capturing diverse visual granularity by constructing and integrating a high-resolution feature pyramid. As a vision-language projector, Hiwin transformer comprises two primary modules: (i) an inverse feature pyramid, constructed by a ViT-derived feature up-sampling process utilizing high-frequency details from an image pyramid, and (ii) hierarchical window attention, focusing on a set of key sampling features within cross-scale windows to condense multi-level feature maps. Extensive experiments demonstrate that LLaVA-UHD v2 achieves superior performance over existing MLLMs on popular benchmarks. Notably, our design brings an average boost of 3.7% across 14 benchmarks compared with the baseline method, 9.3% on DocVQA for instance. We make all the data, model checkpoint, and code publicly available to facilitate future research.

This paper considers image-based virtual try-on, which renders an image of a person wearing a curated garment, given a pair of images depicting the person and the garment, respectively. Previous works adapt existing exemplar-based inpainting diffusion models for virtual try-on to improve the naturalness of the generated visuals compared to other methods (e.g., GAN-based), but they fail to preserve the identity of the garments. To overcome this limitation, we propose a novel diffusion model that improves garment fidelity and generates authentic virtual try-on images. Our method, coined IDM-VTON, uses two different modules to encode the semantics of garment image; given the base UNet of the diffusion model, 1) the high-level semantics extracted from a visual encoder are fused to the cross-attention layer, and then 2) the low-level features extracted from parallel UNet are fused to the self-attention layer. In addition, we provide detailed textual prompts for both garment and person images to enhance the authenticity of the generated visuals. Finally, we present a customization method using a pair of person-garment images, which significantly improves fidelity and authenticity. Our experimental results show that our method outperforms previous approaches (both diffusion-based and GAN-based) in preserving garment details and generating authentic virtual try-on images, both qualitatively and quantitatively. Furthermore, the proposed customization method demonstrates its effectiveness in a real-world scenario.

Recent years have witnessed significant progress in audio-driven human animation. However, critical challenges remain in (i) generating highly dynamic videos while preserving character consistency, (ii) achieving precise emotion alignment between characters and audio, and (iii) enabling multi-character audio-driven animation. To address these challenges, we propose HunyuanVideo-Avatar, a multimodal diffusion transformer (MM-DiT)-based model capable of simultaneously generating dynamic, emotion-controllable, and multi-character dialogue videos. Concretely, HunyuanVideo-Avatar introduces three key innovations: (i) A character image injection module is designed to replace the conventional addition-based character conditioning scheme, eliminating the inherent condition mismatch between training and inference. This ensures the dynamic motion and strong character consistency; (ii) An Audio Emotion Module (AEM) is introduced to extract and transfer the emotional cues from an emotion reference image to the target generated video, enabling fine-grained and accurate emotion style control; (iii) A Face-Aware Audio Adapter (FAA) is proposed to isolate the audio-driven character with latent-level face mask, enabling independent audio injection via cross-attention for multi-character scenarios. These innovations empower HunyuanVideo-Avatar to surpass state-of-the-art methods on benchmark datasets and a newly proposed wild dataset, generating realistic avatars in dynamic, immersive scenarios.

Accurate and interpretable detection of AI-generated images is essential for mitigating risks associated with AI misuse. However, the substantial domain gap among generative models makes it challenging to develop a generalizable forgery detection model. Moreover, since every pixel in an AI-generated image is synthesized, traditional saliency-based forgery explanation methods are not well suited for this task. To address these challenges, we formulate detection and explanation as a unified Forgery Detection and Reasoning task (FDR-Task), leveraging Multi-Modal Large Language Models (MLLMs) to provide accurate detection through reliable reasoning over forgery attributes. To facilitate this task, we introduce the Multi-Modal Forgery Reasoning dataset (MMFR-Dataset), a large-scale dataset containing 120K images across 10 generative models, with 378K reasoning annotations on forgery attributes, enabling comprehensive evaluation of the FDR-Task. Furthermore, we propose FakeReasoning, a forgery detection and reasoning framework with three key components: 1) a dual-branch visual encoder that integrates CLIP and DINO to capture both high-level semantics and low-level artifacts; 2) a Forgery-Aware Feature Fusion Module that leverages DINO's attention maps and cross-attention mechanisms to guide MLLMs toward forgery-related clues; 3) a Classification Probability Mapper that couples language modeling and forgery detection, enhancing overall performance. Experiments across multiple generative models demonstrate that FakeReasoning not only achieves robust generalization but also outperforms state-of-the-art methods on both detection and reasoning tasks.

Medical image segmentation remains particularly challenging for complex and low-contrast anatomical structures. In this paper, we introduce the U-Transformer network, which combines a U-shaped architecture for image segmentation with self- and cross-attention from Transformers. U-Transformer overcomes the inability of U-Nets to model long-range contextual interactions and spatial dependencies, which are arguably crucial for accurate segmentation in challenging contexts. To this end, attention mechanisms are incorporated at two main levels: a self-attention module leverages global interactions between encoder features, while cross-attention in the skip connections allows a fine spatial recovery in the U-Net decoder by filtering out non-semantic features. Experiments on two abdominal CT-image datasets show the large performance gain brought out by U-Transformer compared to U-Net and local Attention U-Nets. We also highlight the importance of using both self- and cross-attention, and the nice interpretability features brought out by U-Transformer.

In this paper, we present SegDINO3D, a novel Transformer encoder-decoder framework for 3D instance segmentation. As 3D training data is generally not as sufficient as 2D training images, SegDINO3D is designed to fully leverage 2D representation from a pre-trained 2D detection model, including both image-level and object-level features, for improving 3D representation. SegDINO3D takes both a point cloud and its associated 2D images as input. In the encoder stage, it first enriches each 3D point by retrieving 2D image features from its corresponding image views and then leverages a 3D encoder for 3D context fusion. In the decoder stage, it formulates 3D object queries as 3D anchor boxes and performs cross-attention from 3D queries to 2D object queries obtained from 2D images using the 2D detection model. These 2D object queries serve as a compact object-level representation of 2D images, effectively avoiding the challenge of keeping thousands of image feature maps in the memory while faithfully preserving the knowledge of the pre-trained 2D model. The introducing of 3D box queries also enables the model to modulate cross-attention using the predicted boxes for more precise querying. SegDINO3D achieves the state-of-the-art performance on the ScanNetV2 and ScanNet200 3D instance segmentation benchmarks. Notably, on the challenging ScanNet200 dataset, SegDINO3D significantly outperforms prior methods by +8.6 and +6.8 mAP on the validation and hidden test sets, respectively, demonstrating its superiority.

Medical Visual Question Answering (Med-VQA) answers clinical questions using medical images, aiding diagnosis. Designing the MedVQA system holds profound importance in assisting clinical diagnosis and enhancing diagnostic accuracy. Building upon this foundation, Hierarchical Medical VQA extends Medical VQA by organizing medical questions into a hierarchical structure and making level-specific predictions to handle fine-grained distinctions. Recently, many studies have proposed hierarchical MedVQA tasks and established datasets, However, several issues still remain: (1) imperfect hierarchical modeling leads to poor differentiation between question levels causing semantic fragmentation across hierarchies. (2) Excessive reliance on implicit learning in Transformer-based cross-modal self-attention fusion methods, which obscures crucial local semantic correlations in medical scenarios. To address these issues, this study proposes a HiCA-VQA method, including two modules: Hierarchical Prompting for fine-grained medical questions and Hierarchical Answer Decoders. The hierarchical prompting module pre-aligns hierarchical text prompts with image features to guide the model in focusing on specific image regions according to question types, while the hierarchical decoder performs separate predictions for questions at different levels to improve accuracy across granularities. The framework also incorporates a cross-attention fusion module where images serve as queries and text as key-value pairs. Experiments on the Rad-Restruct benchmark demonstrate that the HiCA-VQA framework better outperforms existing state-of-the-art methods in answering hierarchical fine-grained questions. This study provides an effective pathway for hierarchical visual question answering systems, advancing medical image understanding.