论文检索

Zhuojiang Cai, Zhenghui Sun, Feng Lu

We present GazeOnce360, a novel end-to-end model for multi-person gaze estimation from a single tabletop-mounted upward-facing fisheye camera. Unlike conventional approaches that rely on forward-facing cameras in constrained viewpoints, we address the underexplored setting of estimating the 3D gaze direction of multiple people distributed across a 360deg scene from an upward fisheye perspective. To support research in this setting, we introduce MPSGaze360, a large-scale synthetic dataset rendered using Unreal Engine, featuring diverse multi-person configurations with accurate 3D gaze and eye landmark annotations. Our model tackles the severe distortion and perspective variation inherent in fisheye imagery by incorporating rotational convolutions and eye landmark supervision. To better capture fine-grained eye features crucial for gaze estimation, we propose a dual-resolution architecture that fuses global low-resolution context with high-resolution local eye regions. Experimental results demonstrate the effectiveness of each component in our model. This work highlights the feasibility and potential of fisheye-based 360deg gaze estimation in practical multi-person scenarios. Project page: https://caizhuojiang.github.io/GazeOnce360/.

Sitong Gong, Yunzhi Zhuge, Lu Zhang, Jiazuo Yu, Pingping Zhang, Xu Jia, Huchuan Lu

Video reasoning segmentation (VRS) endeavors to delineate referred objects in videos guided by implicit instructions that encapsulate human intent and temporal logic. Previous approaches leverage large vision language models (LVLMs) to encode object semantics into \SEG tokens for mask prediction. However, this paradigm suffers from limited interpretability during inference and suboptimal performance due to inadequate spatiotemporal reasoning. Drawing inspiration from seminal breakthroughs in reinforcement learning, we introduce Veason-R1, a specialized LVLM for VRS that emphasizes structured reasoning in segmentation. Veason-R1 is trained through Group Relative Policy Optimization (GRPO) augmented with Chain-of-Thought (CoT) initialization. To begin with, we curate high-quality CoT training data to instill structured reasoning trajectories, bridging video-level semantics and frame-level spatial grounding, yielding the supervised fine-tuned model Veason-SFT. Subsequently, GRPO fine-tuning encourages efficient exploration of the reasoning space by optimizing reasoning chains. To this end, we incorporate a holistic reward mechanism that synergistically enhances spatial alignment and temporal consistency, bolstering keyframe localization and fine-grained grounding. Comprehensive empirical evaluations demonstrate that Veason-R1 achieves state-of-the-art performance on multiple benchmarks, surpassing prior art by significant margins (e.g., +1.3 \mathcal J &\mathcal F in ReVOS and +10.0 \mathcal J &\mathcal F in ReasonVOS), while exhibiting robustness to hallucinations (+8.8 \mathcal R ).

Yizheng Gong, Siyue Yu, Waleed Al-Nuaimy, Jimin Xiao

Vision-language models like CLIP excel at zero-shot recognition but struggle with continual learning due to two critical issues: (1) severe distribution gap between pretraining captions and post-training class names, and (2) performance mismatch between vision-only and dual-encoder approaches--vision-only methods achieve 20% higher accuracy on fine-grained tasks while CLIP dominates on natural images. We propose Learning from Itself (LfI), which mines CLIP's internal knowledge to address both challenges. First, we generate pseudo-captions by optimizing learnable tokens to minimize CLIP's contrastive loss, creating auxiliary training signals that bridge the pretraining-finetuning distribution gap without external models. Second, we introduce adaptive mutual distillation that dynamically weights knowledge transfer between CLIP's text encoder and a temporary vision classifier based on their instantaneous performance--stronger branches teach more, weaker ones learn more. At inference, only the original CLIP architecture is used, having absorbed discriminative knowledge from both branches. LfI achieves state-of-the-art results across multiple continual learning benchmarks, demonstrating that CLIP can effectively teach itself to continually learn new tasks. Code is available at: https://github.com/jordangong/continual-learning.

Yi Feng, Junwu E, Zizhan Guo, Yu Ma, Hanli Wang, Rui Fan

Panoptic occupancy prediction aims to jointly infer voxel-wise semantics and instance identities within a unified 3D scene representation. Nevertheless, progress in this field remains constrained by the absence of high-quality 3D mesh resources, instance-level annotations, and physically consistent occupancy datasets. Existing benchmarks typically provide incomplete and low-resolution geometry without instance-level annotations, limiting the development of models capable of achieving precise geometric reconstruction, reliable occlusion reasoning, and holistic 3D understanding. To address these challenges, this paper presents an instance-centric benchmark for the 3D panoptic occupancy prediction task. Specifically, we introduce ADMesh, the first unified 3D mesh library tailored for autonomous driving, which integrates over 15K high-quality 3D models with diverse textures and rich semantic annotations. Building upon ADMesh, we further construct CarlaOcc, a large-scale, physically consistent panoptic occupancy dataset generated using the CARLA simulator. This dataset contains over 100K frames with fine-grained, instance-level occupancy ground truth at voxel resolutions as fine as 0.05 m. Furthermore, standardized evaluation metrics are introduced to quantify the quality of existing occupancy datasets. Finally, a systematic benchmark of representative models is established on the proposed dataset, which provides a unified platform for fair comparison and reproducible research in the field of 3D panoptic perception. Code and dataset are available at https://mias.group/CarlaOcc.

Jiale Shi, Jiarui Hu, Zesong Yang, Kaixuan Luan, Hujun Bao, Zhaopeng Cui

We introduce GaussianZoom, a generative zoom-in 3D reconstruction system with an iterative progressive framework that combines geometry-consistent scene modeling and multi-scale semantic reasoning to enable high-fidelity extreme zoom-in rendering from low-resolution inputs. To achieve this, we develop a novel multi-view consistent super-resolution module with depth-based feature warping and VLM-driven detail synthesis, ensuring accurate multi-view correspondence while enriching fine-scale appearance beyond the observed resolution. To support zooming across large magnification ranges, we further introduce a new expandable continuous Level-of-Detail hierarchy that dynamically modulates Gaussian visibility for smooth, alias-free cross-scale rendering. Experiments on Mip-NeRF360 and Tanks&Temples demonstrate that GaussianZoom achieves superior perceptual quality, multi-view consistency, and robustness under extreme magnification, establishing a strong baseline for generative zoom-in 3D scene reconstruction.

Shiwei Ren, Tianci Wen, Yongchun Fang, Biao Lu

The dominant 3D Gaussian splatting (3DGS) acceleration methods fail to properly regulate the number of Gaussians during training, causing redundant computational time overhead. In this paper, we propose FastGS, a novel, simple, and general acceleration framework that fully considers the importance of each Gaussian based on multi-view consistency, efficiently solving the trade-off between training time and rendering quality. We innovatively design a densification and pruning strategy based on multi-view consistency, dispensing with the budgeting mechanism. Extensive experiments on Mip-NeRF 360, Tanks & Temples, and Deep Blending datasets demonstrate that our method significantly outperforms the state-of-the-art methods in training speed, achieving a 3.29x training acceleration and comparable rendering quality compared with DashGaussian on the Mip-NeRF 360 dataset and a 15.45x acceleration compared with vanilla 3DGS on the Deep Blending dataset. We demonstrate that FastGS exhibits strong generality, delivering 2-6x training acceleration across various tasks, including dynamic scene reconstruction, surface reconstruction, sparse-view reconstruction, large-scale reconstruction, and simultaneous localization and mapping.

Weiying Xie, Xiaoyu Chen, Xin Zhang, Chenhe Hao, Jitao Ma, Yunsong Li, Leyuan Fang

Vision Transformers (ViTs) exhibit robust performance across diverse visual scenarios. However, their efficiency is constrained by excessive token counts. Token merging offers a viable solution for achieving efficient ViTs. Existing methods merge tokens based solely on specific characteristics within the attention mechanism, which changes significantly across layers. In this paper, we propose a novel training-free SAliency-Driven Token Merging (SAD-TM) approach by leveraging not only the semantic relevance in the attention space but also the latent visual saliency of input patches. Our SAD-TM is inspired by the discovery that saliency-based statistics can directly capture the causal relationship between model input and output, regardless of the layers. Based on the observation, we develop a method that is mathematically formulated to merge tokens with high saliency outliers. The principle behind our merging is that tokens with high saliency outliers usually imply inconsistencies with the global gradient direction, and thus can be merged safely. Besides, our systematic analysis indicates that class attention shows considerable variation across early blocks, so a deferred merging strategy is introduced to optimize the selection of merging rates. In a training-free manner, SAD-TM demonstrates superior performance across various ViT architectures. Especially, with a FLOPs compression of 23.08% on DeiT-Tiny, SAD-TM achieves a Top-1 Accuracy comparable to the pretrained baseline on ImageNet dataset. The code will be available soon.

Siqi Liu, Xinyang Li, Bochao Zou, Junbao Zhuo, Huimin Ma, Jiansheng Chen

As large language models (LLMs) continue to advance, there is increasing interest in their ability to infer human mental states and demonstrate a human-like Theory of Mind (ToM). Most existing ToM evaluations, however, are centered on text-based inputs, while scenarios relying solely on visual information receive far less attention. This leaves a gap, since real-world human-AI interaction typically requires multimodal understanding. In addition, many current methods regard the model as a black box and rarely probe how its internal attention behaves in multiple-choice question answering (QA). The impact of LLM hallucinations on such tasks is also underexplored from an interpretability perspective. To address these issues, we introduce VisionToM, a vision-oriented intervention framework designed to strengthen task-aware reasoning. The core idea is to compute intervention vectors that align visual representations with the correct semantic targets, thereby steering the model's attention through different layers of visual features. This guidance reduces the model's reliance on spurious linguistic priors, leading to more reliable multimodal language model (MLLM) outputs and better QA performance. Experiments on the EgoToM benchmark--an egocentric, real-world video dataset for ToM with three multiple-choice QA settings--demonstrate that our method substantially improves the ToM abilities of MLLMs. Furthermore, results on an additional open-ended generation task show that VisionToM enables MLLMs to produce free-form explanations that more accurately capture agents' mental states, pushing machine-human collaboration toward greater alignment.

Yingkai Zhang, Tao Zhang, Jing Nie, Ying Fu

Unregistered hyperspectral image (HSI) super-resolution (SR) typically aims to enhance a low-resolution HSI using an unregistered high-resolution reference image. In this paper, we propose an unmixing-based fusion framework that decouples spatial-spectral information to simultaneously mitigate the impact of unregistered fusion and enhance the learnability of SR models. Specifically, we first utilize singular value decomposition for initial spectral unmixing, preserving the original endmembers while dedicating the subsequent network to enhancing the initial abundance map. To leverage the spatial texture of the unregistered reference, we introduce a coarse-to-fine deformable aggregation module, which first estimates a pixel-level flow and a similarity map using a coarse pyramid predictor. It further performs fine sub-pixel refinement to achieve deformable aggregation of the reference features. The aggregative features are then refined via a series of spatial-channel abundance cross-attention blocks. Furthermore, a spatial-channel modulated fusion module is presented to merge encoder-decoder features using dynamic gating weights, yielding a high-quality, high-resolution HSI. Experimental results on simulated and real datasets confirm that our proposed method achieves state-of-the-art super-resolution performance. The code will be available at https://github.com/yingkai-zhang/UAFL.

Qing Huang, Zhipei Xu, Xuanyu Zhang, Xiangyu Yu, Jian Zhang

The rise of AI-generated images (AIGIs) poses growing challenges for digital authenticity, prompting the need for efficient, generalizable image forgery detection systems. Existing methods, whether non-LLM-based or LLM-based, exhibit distinct advantages and limitations. While non-LLM-based models offer efficient low-level artifact detection, they often lack semantic understanding. Conversely, LLM-based methods provide strong semantic reasoning and explainability but are computationally intensive and less sensitive to subtle visual artifacts. Moreover, the true contribution of explanatory reasoning texts to forgery detection performance remains unclear. In this work, we investigate the intrinsic value and potential of LLM-generated reasoning texts, considering it a source of generalization and semantic-error sensitivity. Based on these findings, we propose ReAlign, a novel framework that distills high-quality reasoning texts generated by a GRPO-optimized LLM into a lightweight AIGI detector via contrastive learning. ReAlign effectively inherits the generalization ability and semantic sensitivity capability of reasoning textual representations, while remaining efficient and lightweight for deployment. Moreover, ReAlign adopts a tailored joint optimization strategy that integrates contrastive loss for image-text alignment and classification loss for accurate forgery discrimination. Experimental results on AIGCDetectBenchmark, AIGI-Holmes, and our newly constructed UltraSynth-10k demonstrate that ReAlign consistently outperforms existing state-of-the-art detectors in both accuracy and generalization, particularly when facing complex, high-fidelity forgeries from modern generative models.

Wangwang Jia, Zijian Gao, Tianjiao Wan, Yuan Cao, Yong Dou, Kele Xu

Class-Incremental Learning (CIL) seeks to acquire new classes over time without erasing prior knowledge. While recent methods leverage pre-trained models (PTMs) to curb forgetting, they largely optimize the feature extractor and overlook the crucial classification head. In this work, we advance a simple view: if each task is equipped with a classifier that has the ability to both recognize in-distribution (IND) classes and reject out-of-distribution (OOD) inputs, CIL arises naturally--inputs are accepted only by heads that deem them in-distribution and rejected otherwise. Supported by rigorous theoretical and empirical studies, we find that this ability is characterized by Inter-class Separation and Intra-class Compactness; lacking these, standard linear and cosine-similarity heads remain closed-set and fail to yield a usable OOD signal. To address this, we propose GOD (Geometry-driven OOD Detectors), which unifies IND recognition and OOD rejection in a single geometric space by replacing the learnable head with fixed Equiangular Tight Frame (ETF) anchors; an ETF loss enforces inter-class separation, and an ArcFace loss further tightens intra-class compactness. For efficiency, we further introduce a parameter-efficient hybrid architecture and an efficient inference strategy, thus reducing both parameter footprint and inference cost. Extensive experiments on multiple incremental settings and datasets show that GOD achieves state-of-the-art results.[^1]: Code and datasets are available in the supplementary material.

Ping Chen, Zezhou Chen, Xingpeng Zhang, Yanlin Qian, Huan Hu, Xiang Liu, Zipeng Wang, Xin Wang, Zhaoxiang Liu, Kai Wang 等

Current 2D-to-3D conversion methods achieve geometric accuracy but are artistically deficient, failing to replicate the immersive and emotionally resonant experience of professional 3D cinema. This is because "geometric reconstruction" paradigms mistake deliberate artistic intent--such as strategic zero-plane shifts for "pop-out" effects and local depth sculpting--for data "noise" or ambiguity. This paper argues for a new paradigm: Artistic Disparity Synthesis, shifting the goal from physically accurate disparity estimation to artistically coherent disparity synthesis. We propose Art3D, a preliminary framework exploring this paradigm. Art3D uses a dual-path architecture to decouple global depth parameters (macro-intent) from local artistic effects (visual brushstrokes) and learns from professional 3D film data via indirect supervision. We also introduce a preliminary evaluation method to quantify cinematic alignment. Experiments show our approach demonstrates potential in replicating key local out-of-screen effects and aligning with the global depth styles of cinematic 3D content, laying the groundwork for a new class of artistically-driven conversion tools.

Zichuan Lin, Yicheng Liu, Yang Yang, Lvfang Tao, Deheng Ye

Vision-Language Models (VLMs) have achieved remarkable success in visual question answering tasks, but their reliance on large numbers of visual tokens introduces significant computational overhead. While existing efficient VLM approaches reduce visual tokens through fixed-ratio compression, they operate passively and lack the ability to adapt to varying task requirements. This motivates a fundamental question: Can VLMs autonomously determine the minimum number of visual tokens required for each sample? Inspired by human active vision mechanisms, we introduce AdaptVision, an efficient VLM paradigm that enables adaptive visual token acquisition through a coarse-to-fine approach. Our model initially processes compressed visual tokens from low-resolution images and selectively acquires additional visual information by invoking a bounding box tool to crop key regions when necessary. We train AdaptVision using a reinforcement learning framework that carefully balances accuracy and efficiency. Central to our approach is Decoupled Turn Policy Optimization (DTPO), which decouples the learning objective into two components: (1) tool learning, which optimizes correct tool utilization, and (2) accuracy improvement, which refines the generated responses to improve answer correctness. Based on this formulation, we further decouple advantage estimation by computing separate advantages for tokens associated with each objective. This formulation enables more effective optimization for AdaptVision compared to vanilla GRPO. Comprehensive experiments across multiple VQA benchmarks demonstrate that AdaptVision achieves superior performance while consuming substantially fewer visual tokens than state-of-the-art efficient VLM methods.

Tong Wei, Giorgos Tolias, Jiri Matas, Daniel Barath

The pose graph is a core component of Structure-from-Motion (SfM), where images act as nodes and edges encode relative poses. Since geometric verification is expensive, SfM pipelines restrict the pose graph to a sparse set of candidate edges, making initialization critical. Existing methods rely on image retrieval to connect each image to its k nearest neighbors, treating pairs independently and ignoring global consistency. We address this limitation through the concept of edge prioritization, ranking candidate edges by their utility for SfM. Our approach has three components: (1) a GNN trained with SfM-derived supervision to predict globally consistent edge reliability; (2) multi-minimal-spanning-tree-based pose graph construction guided by these ranks; and (3) connectivity-aware score modulation that reinforces weak regions and reduces graph diameter. This globally informed initialization yields more reliable and compact pose graphs, improving reconstruction accuracy in sparse and high-speed settings and outperforming SOTA retrieval methods on ambiguous scenes. The code and trained models are available at https://github.com/weitong8591/global_edge_prior.

Jiahao Huang, Fengyan Lin, Xuechao Yang, Chen Feng, Kexin Zhu, Xu Yang, Zhide Chen

The development of affective multimodal language models (MLMs) has long been constrained by a gap between low-level perception and high-level interaction, leading to fragmented affective capabilities and limited generalization. To bridge this gap, we propose a cognitively inspired three-level hierarchy that organizes affective tasks according to their cognitive depth--perception, understanding, and interaction--and provides a unified conceptual foundation for advancing affective modeling. Guided by this hierarchy, we introduce Nano-EmoX, a small-scale multitask MLM, and P2E (Perception-to-Empathy), a curriculum-based training framework. Nano-EmoX integrates a suite of omni-modal encoders, including an enhanced facial encoder and a fusion encoder, to capture key multimodal affective cues and improve cross-task transferability. The outputs are projected into a unified language space via heterogeneous adapters, empowering a lightweight language model to tackle diverse affective tasks. Concurrently, P2E progressively cultivates emotional intelligence by aligning rapid perception with chain-of-thought-driven empathy. To the best of our knowledge, Nano-EmoX is the first compact MLM (2.2B) to unify six core affective tasks across all three hierarchy levels, achieving state-of-the-art or highly competitive performance across multiple benchmarks, demonstrating excellent efficiency and generalization.

Jinge Ma, Fengqing Zhu

With the widespread adoption of deep learning in visual tasks, Class-Incremental Learning (CIL) has become an important paradigm for handling dynamically evolving data distributions. However, CIL faces the core challenge of *catastrophic forgetting*, often manifested as a prediction bias toward new classes. Existing methods mainly attribute this bias to intra-task class imbalance and focus on corrections at the classifier head. In this paper, we highlight an overlooked factor--*temporal imbalance*--as a key cause of this bias. Earlier classes receive stronger negative supervision toward the end of training, leading to asymmetric precision and recall. We establish a temporal supervision model, formally define temporal imbalance, and propose the Temporal-Adjusted Loss (TAL), which uses a temporal decay kernel to construct a supervision strength vector and dynamically reweight the negative supervision in cross-entropy loss. Theoretical analysis shows that TAL degenerates to standard cross-entropy under balanced conditions and effectively mitigates prediction bias under imbalance. Extensive experiments demonstrate that TAL significantly reduces forgetting and improves performance on multiple CIL benchmarks, underscoring the importance of temporal modeling for stable long-term learning.

Yuran Wang, Bohan Zeng, Chengzhuo Tong, Wenxuan Liu, Yang Shi, Xiaochen Ma, Hao Liang, Yuanxing Zhang, Wentao Zhang

Subject-driven image generation has advanced from single- to multi-subject composition, while neglecting distinction, the ability to distinguish and generate the correct subject when inputs contain multiple candidates. This limitation restricts effectiveness in complex, realistic visual settings. We propose Scone, a unified understanding-generation method that integrates composition and distinction. Scone enables the understanding expert to act as a semantic bridge, conveying semantic information and guiding the generation expert to preserve subject identity while minimizing interference. A two-stage training scheme first learns composition, then enhances distinction through semantic alignment and attention-based masking. We also introduce SconeEval, a benchmark for evaluating both composition and distinction across diverse scenarios. Experiments demonstrate that Scone outperforms existing open-source models in composition and distinction tasks on two benchmarks. Our model, benchmark, and training data are available at: https://github.com/Ryann-Ran/Scone.

Anqi Zhang, Xiaokang Ji, Guangyu Gao, Jianbo Jiao, Chi Harold Liu, Yunchao Wei

Recent segmentation methods leveraging Multi-modal Large Language Models (MLLMs) have shown reliable object-level segmentation and enhanced spatial perception. However, almost all previous methods predominantly rely on specialist mask decoders to interpret masks from generated segmentation-related embeddings and visual features, or incorporate multiple additional tokens to assist. This paper aims to investigate whether and how we can unlock segmentation from MLLM itSELF with 1 segmentation Embedding (SELF1E) while achieving competitive results, which eliminates the need for external decoders. To this end, our approach targets the fundamental limitation of resolution reduction in pixel-shuffled image features from MLLMs. First, we retain image features at their original uncompressed resolution, and refill them with residual features extracted from MLLM-processed compressed features, thereby improving feature precision. Subsequently, we integrate pixel-unshuffle operations on image features with and without LLM processing, respectively, to unleash the details of compressed features and amplify the residual features under uncompressed resolution, which further enhances the resolution of refilled features. Moreover, we redesign the attention mask with dual perception pathways, i.e., image-to-image and image-to-segmentation, enabling rich feature interaction between pixels and the segmentation token. Comprehensive experiments across multiple segmentation tasks validate that SELF1E achieves performance competitive with specialist mask decoder-based methods, demonstrating the feasibility of decoder-free segmentation in MLLMs. Project page: https://github.com/ANDYZAQ/SELF1E.

Matthew Walmer, Saksham Suri, Anirud Aggarwal, Abhinav Shrivastava

The space of task-agnostic feature upsampling has emerged as a promising area of research to efficiently create denser features from pre-trained visual backbones. These methods act as a shortcut to achieve dense features for a fraction of the cost by learning to map low-resolution features to high-resolution versions. While early works in this space used iterative upsampling approaches, more recent works have switched to cross-attention-based methods, which risk falling into the same efficiency scaling problems of the backbones they are upsampling. In this work, we demonstrate that iterative upsampling methods can still compete with cross-attention-based methods; moreover, they can achieve state-of-the-art performance with lower inference costs. We propose UPLiFT, an architecture for Universal Pixel-dense Lightweight Feature Transforms. We also propose an efficient Local Attender operator to overcome the limitations of prior iterative feature upsampling methods. This operator uses an alternative attentional pooling formulation defined fully locally. We show that our Local Attender allows UPLiFT to maintain stable features throughout upsampling, enabling state-of-the-art performance with lower inference costs than existing pixel-dense feature upsamplers. In addition, we apply UPLiFT to generative downstream tasks and show that it achieves competitive performance with state-of-the-art Coupled Flow Matching models for VAE feature upsampling. Altogether, UPLiFT offers a versatile and efficient approach to creating denser features.

Yuxin Yang, Yinan Zhou, Yuxin Chen, Ziqi Zhang, Zongyang Ma, Chunfeng Yuan, Bing Li, Jun Gao, Weiming Hu

Composed Image Retrieval (CIR) has demonstrated significant potential by enabling flexible, multimodal queries that combine a reference image and modification text. However, CIR inherently prioritizes semantic matching, struggling to reliably retrieve a user-specified instance across contexts. In practice, emphasizing concrete instance fidelity over broad semantics is often more consequential. In this work, we propose Object-Anchored Composed Image Retrieval (OACIR), a novel fine-grained retrieval task that mandates strict instance-level consistency. To advance research on this task, we construct OACIRR (OACIR on Real-world images), the first large-scale, multi-domain benchmark comprising over 160K quadruples and four challenging candidate galleries enriched with hard-negative instance distractors. Each quadruple augments the compositional query with a bounding box that visually anchors the object in the reference image, providing a precise and flexible way to ensure instance preservation. To address the OACIR task, we propose AdaFocal, a framework featuring a Context-Aware Attention Modulator that adaptively intensifies attention within the specified instance region, dynamically balancing focus between the anchored instance and the broader compositional context. Extensive experiments demonstrate that AdaFocal substantially outperforms existing compositional retrieval models, particularly in maintaining instance-level fidelity, thereby establishing a robust baseline for this challenging task while opening new directions for more flexible, instance-aware retrieval systems.