Vision-Language Alignment

High text recognition performance does not guarantee that Vision-Language Models (VLMs) share human-like decision patterns when resolving ambiguity. We investigate this behavioral gap by directly comparing humans and VLMs using continuously interpolated Japanese character shapes generated via a $β$-VAE. We estimate decision boundaries in a single-character recognition (shape-only task) and evaluate whether VLM responses align with human judgments under shape in context (i.e., embedding an ambiguous character near the human decision boundary in word-level context). We find that human and VLM decision boundaries differ in the shape-only task, and that shape in context can improve human alignment in some conditions. These results highlight qualitative behavioral differences, offering foundational insights toward human--VLM alignment benchmarking.

Large unimodal foundation models for vision and language encode rich semantic structures, yet aligning them typically requires computationally intensive multimodal fine-tuning. Such approaches depend on large-scale parameter updates, are resource intensive, and can perturb pretrained representations. Emerging evidence suggests, however, that independently trained foundation models may already exhibit latent semantic compatibility, reflecting shared structures in the data they model. This raises a fundamental question: can cross-modal alignment be achieved without modifying the models themselves? Here we introduce HDFLIM (HyperDimensional computing with Frozen Language and Image Models), a framework that establishes cross-modal mappings while keeping pretrained vision and language models fully frozen. HDFLIM projects unimodal embeddings into a shared hyperdimensional space and leverages lightweight symbolic operations -- binding, bundling, and similarity-based retrieval to construct associative cross-modal representations in a single pass over the data. Caption generation emerges from high-dimensional memory retrieval rather than iterative gradient-based optimization. We show that HDFLIM achieves performance comparable to end-to-end vision-language training methods and produces captions that are more semantically grounded than zero-shot baselines. By decoupling alignment from parameter tuning, our results suggest that semantic mapping across foundation models can be realized through symbolic operations on hyperdimensional encodings of the respective embeddings. More broadly, this work points toward an alternative paradigm for foundation model alignment in which frozen models are integrated through structured representational mappings rather than through large-scale retraining. The codebase for our implementation can be found at https://github.com/Abhishek-Dalvi410/HDFLIM.

The Platonic Representation Hypothesis posits that neural networks trained on different modalities converge toward a shared statistical model of the world. Recent work exploits this convergence by aligning frozen pretrained vision and language models with lightweight alignment layers, but typically relies on contrastive losses and millions of paired samples. In this work, we ask whether meaningful alignment can be achieved with substantially less supervision. We introduce a semi-supervised setting in which pretrained unimodal encoders are aligned using a small number of image-text pairs together with large amounts of unpaired data. To address this challenge, we propose SOTAlign, a two-stage framework that first recovers a coarse shared geometry from limited paired data using a linear teacher, and then refines the alignment on unpaired samples via an optimal-transport-based divergence that transfers relational structure without overconstraining the target space. SOTAlign effectively leverages unpaired images and text, learning robust joint embeddings across datasets and encoder pairs, and significantly outperforming supervised and semi-supervised baselines. Code is available at https://github.com/ExplainableML/SOTAlign.

Referring Image Segmentation (RIS) aims to segment the object in an image uniquely referred to by a natural language expression. However, RIS training often contains hard-to-align and instance-specific visual signals; optimizing on such pixels injects misleading gradients and drives the model in the wrong direction. By explicitly estimating pixel-level vision-language alignment, the learner can suppress low-alignment regions, concentrate on reliable cues, and acquire more generalizable alignment features. In this paper, we propose Alignment-Aware Masked Learning (AML), a simple yet effective training strategy that quantifies region-referent alignment (PMME) and filters out unreliable pixels during optimization (AFM). Specifically, each sample first computes a similarity map between visual and textual features, and then masks out pixels falling below an adaptive similarity threshold, thereby excluding poorly aligned regions from the training process. AML does not require architectural changes and incurs no inference overhead, directing attention to the areas aligned with the textual description. Experiments on the RefCOCO (vanilla/+/g) datasets show that AML achieves state-of-the-art results across all 8 splits, and beyond improving RIS performance, AML also enhances the model's robustness to diverse descriptions and scenarios. Code is available at https://github.com/pipashu1/AMLRIS.

Image-text retrieval has become a fundamental component in intelligent multimedia systems; however, most existing vision-language models are optimized for highresource languages and remain suboptimal for low-resource settings such as Vietnamese. This work introduces ViCLIP-OT, a foundation vision-language model specifically designed for Vietnamese image-text retrieval. The proposed framework integrates CLIP-style contrastive learning with a Similarity-Graph Regularized Optimal Transport (SIGROT) loss to enhance global cross-modal consistency and mitigate modality gap issues. Extensive experiments on three Vietnamese benchmarks (UITOpenViIC, KTVIC, and Crossmodal-3600) demonstrate that ViCLIP-OT consistently outperforms CLIP and SigLIP baselines in both in-domain and zero-shot settings. On UIT-OpenViIC, the model achieves an average Recall@K of 67.34%, improving upon CLIP by 5.75 percentage points. In zero-shot evaluation on Crossmodal-3600, ViCLIPOT surpasses CLIP by 11.72 percentage points. Embedding-space analysis further confirms improved alignment and reduced modality gap. The results indicate that integrating SIGROT provides an effective and scalable strategy for cross-modal retrieval in low-resource languages, offering practical implications for intelligent multimedia retrieval systems in Vietnamese and other underrepresented linguistic contexts.

Vision-language foundation models (VLFMs) promise zero-shot and retrieval understanding for Earth observation. While operational satellite systems often lack full multi-spectral coverage, making RGB-only inference highly desirable for scalable deployment, the adoption of VLFMs for satellite imagery remains hindered by two factors: (1) multi-spectral inputs are informative but difficult to exploit consistently due to band redundancy and misalignment; and (2) CLIP-style text encoders limit semantic expressiveness and weaken fine-grained alignment. We present SATtxt, a spectrum-aware VLFM that operates with RGB inputs only at inference while retaining spectral cues learned during training. Our framework comprises two stages. First, Spectral Representation Distillation transfers spectral priors from a frozen multi-spectral teacher to an RGB student via a lightweight projector. Second, Spectrally Grounded Alignment with Instruction-Augmented LLMs bridges the distilled visual space and an expressive LLM embedding space. Across EuroSAT, BigEarthNet, and ForestNet, SATtxt improves zero-shot classification on average by 4.2%, retrieval by 5.9%, and linear probing by 2.7% over baselines, showing an efficient path toward spectrum-aware vision-language learning for Earth observation. Project page: https://ikhado.github.io/sattxt/

CLIP models learn transferable multi-modal features via image-text contrastive learning on internet-scale data. They are widely used in zero-shot classification, multi-modal retrieval, text-to-image diffusion, and as image encoders in large vision-language models. However, CLIP's pretraining is dominated by images paired with short captions, biasing the model toward encoding simple descriptions of salient objects and leading to coarse alignment on complex scenes and dense descriptions. While recent work mitigates this by fine-tuning on small-scale long-caption datasets, we identify an important common bias: both human- and LLM-generated long captions typically begin with a one-sentence summary followed by a detailed description. We show that this acts as a shortcut during training, concentrating attention on the opening sentence and early tokens and weakening alignment over the rest of the caption. To resolve this, we introduce DeBias-CLIP, which removes the summary sentence during training and applies sentence sub-sampling and text token padding to distribute supervision across all token positions. DeBias-CLIP achieves state-of-the-art long-text retrieval, improves short-text retrieval, and is less sensitive to sentence order permutations. It is a drop-in replacement for Long-CLIP with no additional trainable parameters.

Visual storytelling models that correctly ground entities in images may still hallucinate semantic relationships, generating incorrect dialogue attribution, character interactions, or emotional states. We introduce StoryMovie, a dataset of 1,757 stories aligned with movie scripts and subtitles through LCS matching. Our alignment pipeline synchronizes screenplay dialogue with subtitle timestamps, enabling dialogue attribution by linking character names from scripts to temporal positions from subtitles. Using this aligned content, we generate stories that maintain visual grounding tags while incorporating authentic character names, dialogue, and relationship dynamics. We fine-tune Qwen Storyteller3 on this dataset, building on prior work in visual grounding and entity re-identification. Evaluation using DeepSeek V3 as judge shows that Storyteller3 achieves an 89.9% win rate against base Qwen2.5-VL 7B on subtitle alignment. Compared to Storyteller, trained without script grounding, Storyteller3 achieves 48.5% versus 38.0%, confirming that semantic alignment progressively improves dialogue attribution beyond visual grounding alone.

Prompt learning has become a dominant paradigm for adapting vision-language models (VLMs) such as CLIP to downstream tasks without modifying pretrained weights. While extending prompts to both vision and text encoders across multiple transformer layers significantly boosts performance, it dramatically increases the number of trainable parameters, with state-of-the-art methods requiring millions of parameters and abandoning the parameter efficiency that makes prompt tuning attractive. In this work, we propose MMLoP (Multi-Modal Low-Rank Prompting), a framework that achieves deep multi-modal prompting with only 11.5K trainable parameters, comparable to early text-only methods like CoOp. MMLoP parameterizes vision and text prompts at each transformer layer through a low-rank factorization that constrains prompts to a compact subspace, providing parameter efficiency while motivating the need for our complementary regularization components. To further close the accuracy gap with state-of-the-art methods, we introduce three complementary components: a self-regulating consistency loss that anchors prompted representations to frozen zero-shot CLIP features at both the feature and logit levels, a uniform drift correction that removes the global embedding shift induced by prompt tuning to preserve class-discriminative structure, and a shared up-projection that couples vision and text prompts through a common low-rank factor to enforce cross-modal alignment. Extensive experiments across three benchmarks and 11 diverse datasets demonstrate that MMLoP achieves a highly favorable accuracy-efficiency tradeoff, outperforming the majority of existing methods including those with orders of magnitude more parameters, while achieving a harmonic mean of 79.70% on base-to-novel generalization. Code is available at https://github.com/sajjad-ucsb/MMLoP.

Vision-Language Models (VLMs) like CLIP struggle to understand negation, often embedding affirmatives and negatives similarly (e.g., matching "no dog" with dog images). Existing methods refine negation understanding via fine-tuning CLIP's text encoder, risking overfitting. In this work, we propose CLIPGlasses, a plug-and-play framework that enhances CLIP's ability to comprehend negated visual descriptions. CLIPGlasses adopts a dual-stage design: a Lens module disentangles negated semantics from text embeddings, and a Frame module predicts context-aware repulsion strength, which is integrated into a modified similarity computation to penalize alignment with negated semantics, thereby reducing false positive matches. Experiments show that CLIP equipped with CLIPGlasses achieves competitive in-domain performance and outperforms state-of-the-art methods in cross-domain generalization. Its superiority is especially evident under low-resource conditions, indicating stronger robustness across domains.

Edge-based representations are fundamental cues for visual understanding, a principle rooted in early vision research and still central today. We extend this principle to vision-language alignment, showing that isolating and aligning structural cues across modalities can greatly benefit fine-tuning on long, detail-rich captions, with a specific focus on improving cross-modal retrieval. We introduce StructXLIP, a fine-tuning alignment paradigm that extracts edge maps (e.g., Canny), treating them as proxies for the visual structure of an image, and filters the corresponding captions to emphasize structural cues, making them "structure-centric". Fine-tuning augments the standard alignment loss with three structure-centric losses: (i) aligning edge maps with structural text, (ii) matching local edge regions to textual chunks, and (iii) connecting edge maps to color images to prevent representation drift. From a theoretical standpoint, while standard CLIP maximizes the mutual information between visual and textual embeddings, StructXLIP additionally maximizes the mutual information between multimodal structural representations. This auxiliary optimization is intrinsically harder, guiding the model toward more robust and semantically stable minima, enhancing vision-language alignment. Beyond outperforming current competitors on cross-modal retrieval in both general and specialized domains, our method serves as a general boosting recipe that can be integrated into future approaches in a plug-and-play manner. Code and pretrained models are publicly available at: https://github.com/intelligolabs/StructXLIP.

In multimodal learning, CLIP has emerged as the de-facto approach for mapping different modalities into a shared latent space by bringing semantically similar representations closer while pushing apart dissimilar ones. However, CLIP-based contrastive losses exhibit unintended behaviors that negatively impact true semantic alignment, leading to sparse and fragmented latent spaces. This phenomenon, known as the modality gap, has been partially mitigated for standard text and image pairs but remains unknown and unresolved in more complex multimodal settings, such as the medical domain. In this work, we study this phenomenon in the latter case, revealing that the modality gap is present also in medical alignment, and we propose a modality-agnostic framework that closes this gap, ensuring that semantically related representations are more aligned, regardless of their source modality. Our method enhances alignment between radiology images and clinical text, improving cross-modal retrieval and image captioning.

Vision-language models (VLMs) can describe urban scenes in rich detail, yet consistently fail to produce reliable human preference labels in domain-specific tasks such as safety assessment and aesthetic evaluation. The standard fix, fine-tuning or RLHF, requires large-scale annotations and model retraining. We ask a different question: can a frozen VLM be aligned with human preferences without modifying any weights? Our key insight is that VLMs are strong concept extractors but poor decision calibrators. We propose a three-stage post-hoc pipeline that exploits this asymmetry: (i) interpretable evaluation dimensions are automatically mined from consensus exemplars; (ii) an Observer-Debater-Judge chain extracts robust concept scores from the frozen VLM; and (iii) locally-weighted ridge regression on a hybrid manifold calibrates these scores to human ratings. Applied as UrbanAlign on Place Pulse 2.0, the framework reaches 72.2% accuracy (kappa=0.45) across six perception categories, outperforming all baselines by +11.0 pp and zero-shot VLM by +15.5 pp, with full interpretability and zero weight modification.

Vision-Language-Action (VLA) models enable instruction-following robotic manipulation, but they are typically pretrained on 2D data and lack 3D spatial understanding. An effective approach is representation alignment, where a strong vision foundation model is used to guide a 2D VLA model. However, existing methods usually apply supervision at only a single layer, failing to fully exploit the rich information distributed across depth; meanwhile, naïve multi-layer alignment can cause gradient interference. We introduce ROCKET, a residual-oriented multi-layer representation alignment framework that formulates multi-layer alignment as aligning one residual stream to another. Concretely, ROCKET employs a shared projector to align multiple layers of the VLA backbone with multiple layers of a powerful 3D vision foundation model via a layer-invariant mapping, which reduces gradient conflicts. We provide both theoretical justification and empirical analyses showing that a shared projector is sufficient and outperforms prior designs, and further propose a Matryoshka-style sparse activation scheme for the shared projector to balance multiple alignment losses. Our experiments show that, combined with a training-free layer selection strategy, ROCKET requires only about 4% of the compute budget while achieving 98.5% state-of-the-art success rate on LIBERO. We further demonstrate the superior performance of ROCKET across LIBERO-Plus and RoboTwin, as well as multiple VLA models. The code and model weights can be found at https://github.com/CASE-Lab-UMD/ROCKET-VLA.

Generative models have made it possible to synthesize highly realistic images, potentially providing an abundant data source for training machine learning models. Despite the advantages of these synthesizable data sources, the indiscriminate use of generated images as real images for training can even cause mode collapse due to modality discrepancies between real and synthetic domains. In this paper, we propose a novel framework for discriminative use of generated images, coined GMAIL, that explicitly treats generated images as a separate modality from real images. Instead of indiscriminately replacing real images with generated ones in the pixel space, our approach bridges the two distinct modalities in the same latent space through a multi-modal learning approach. To be specific, we first fine-tune a model exclusively on generated images using a cross-modality alignment loss and then employ this aligned model to further train various vision-language models with generated images. By aligning the two modalities, our approach effectively leverages the benefits of recent advances in generative models, thereby boosting the effectiveness of generated image learning across a range of vision-language tasks. Our framework can be easily incorporated with various vision-language models, and we demonstrate its efficacy throughout extensive experiments. For example, our framework significantly improves performance on image captioning, zero-shot image retrieval, zero-shot image classification, and long caption retrieval tasks. It also shows positive generated data scaling trends and notable enhancements in the captioning performance of the large multimodal model, LLaVA.

Referring Image Segmentation (RIS) aims to segment a target object described by a natural language expression. Existing methods have evolved by leveraging the vision information into the language tokens. To more effectively exploit visual contexts for fine-grained segmentation, we propose a novel Visual Informative Part Attention (VIPA) framework for referring image segmentation. VIPA leverages the informative parts of visual contexts, called a visual expression, which can effectively provide the structural and semantic visual target information to the network. This design reduces high-variance cross-modal projection and enhances semantic consistency in an attention mechanism of the referring image segmentation. We also design a visual expression generator (VEG) module, which retrieves informative visual tokens via local-global linguistic context cues and refines the retrieved tokens for reducing noise information and sharing informative visual attributes. This module allows the visual expression to consider comprehensive contexts and capture semantic visual contexts of informative regions. In this way, our framework enables the network's attention to robustly align with the fine-grained regions of interest. Extensive experiments and visual analysis demonstrate the effectiveness of our approach. Our VIPA outperforms the existing state-of-the-art methods on four public RIS benchmarks.

Radiology Report Generation aims to produce accurate and coherent diagnostics from medical images. Although large vision-language models improve report fluency and accuracy, they still suffer from hallucinations by generating plausible pathological descriptions that are not supported by the input images. Existing methods primarily rely on external knowledge guidance to facilitate the alignment between generated text and visual information. However, these approaches often ignore the inherent decoding priors and vision-language alignment biases in pretrained models and lack robustness due to reliance on constructed guidance. In this paper, we propose Layer-wise Expert-aligned Decoding, a method that directly intervenes in the internal decoding process of large vision-language models. A pathology-specific expert module is designed to extract discriminative pathological features, which are then injected into each decoder layer through a gated mechanism. This architecture enables the large language model to progressively incorporate expert features during generation through a learned layer-wise gating function, thereby mitigating decoding biases and steering generation toward factual consistency. Experiments on multiple public datasets demonstrate that the proposed method improves clinical accuracy and factual consistency while maintaining competitive report generation quality.

Text-to-video (T2V) generation aims to synthesize videos with high visual quality and temporal consistency that are semantically aligned with input text. Reward-based post-training has emerged as a promising direction to improve the quality and semantic alignment of generated videos. However, recent methods either rely on large-scale human preference annotations or operate on misaligned embeddings from pre-trained vision-language models, leading to limited scalability or suboptimal supervision. We present $\texttt{PISCES}$, an annotation-free post-training algorithm that addresses these limitations via a novel Dual Optimal Transport (OT)-aligned Rewards module. To align reward signals with human judgment, $\texttt{PISCES}$ uses OT to bridge text and video embeddings at both distributional and discrete token levels, enabling reward supervision to fulfill two objectives: (i) a Distributional OT-aligned Quality Reward that captures overall visual quality and temporal coherence; and (ii) a Discrete Token-level OT-aligned Semantic Reward that enforces semantic, spatio-temporal correspondence between text and video tokens. To our knowledge, $\texttt{PISCES}$ is the first to improve annotation-free reward supervision in generative post-training through the lens of OT. Experiments on both short- and long-video generation show that $\texttt{PISCES}$ outperforms both annotation-based and annotation-free methods on VBench across Quality and Semantic scores, with human preference studies further validating its effectiveness. We show that the Dual OT-aligned Rewards module is compatible with multiple optimization paradigms, including direct backpropagation and reinforcement learning fine-tuning. Project page: https://roar-ai.github.io/pisces

Deepfake detection is a widely researched topic that is crucial for combating the spread of malicious content, with existing methods mainly modeling the problem as classification or spatial localization. The rapid advancements in generative models impose new demands on Deepfake detection. In this paper, we propose multimodal alignment and reinforcement for explainable Deepfake detection via vision-language models, termed MARE, which aims to enhance the accuracy and reliability of Vision-Language Models (VLMs) in Deepfake detection and reasoning. Specifically, MARE designs comprehensive reward functions, incorporating reinforcement learning from human feedback (RLHF), to incentivize the generation of text-spatially aligned reasoning content that adheres to human preferences. Besides, MARE introduces a forgery disentanglement module to capture intrinsic forgery traces from high-level facial semantics, thereby improving its authenticity detection capability. We conduct thorough evaluations on the reasoning content generated by MARE. Both quantitative and qualitative experimental results demonstrate that MARE achieves state-of-the-art performance in terms of accuracy and reliability.

Conditional diffusion models frequently suffer from language-image misalignments. Due to the ambiguity of intermediate noise corrupted latents, assessing prompt adherence currently requires completing the entire sampling trajectory. This late-stage evaluation incurs even higher computational costs during test-time scaling strategies, such as Best-of-N (BoN) sampling, as all misaligned trajectories must finish generation before being discarded. To tackle this, we propose NoisyCLIP, a noise-aware twin-tower model that enables early language-to-latent alignment estimation. By learning a vision encoder on noise-corrupted latents, we allow the model to "see" through the ambiguity of intermediate diffusion steps. To facilitate this training, we investigate noise-data augmentation sampling strategies and introduce two new benchmark datasets: Noisy-Conceptual-Captions and Noisy-GenAI-Bench. When applied as an early-stopping criterion for BoN, NoisyCLIP at half cost matches or beats frozen CLIP at full cost. Ultimately, this transforms alignment assessment from an expensive final check into a continuous monitoring tool, drastically reducing compute costs without sacrificing semantic fidelity.

In the paradigm of computer vision-based precise real-world scene understanding, joint reasoning in terms of contextual understanding about the objects present in a scene, their inter-object relations, and the action being performed is an essential prerequisite. However, prior works have not addressed all three jointly, and no large-scale dataset provides grounded annotations at all three levels across diverse visual scenarios. Hence, this work introduces the SCLARO (Scene-Contextual Localisation of Actions, Relations & Objects) dataset, consisting of 615,805 images spanning indoor, outdoor, and driving scenarios, annotated with global action captions, object bounding boxes, and relation triplets that supply structured scene context beyond a free-text caption. To benchmark the dataset, we propose ScenarioCLIP, a tri-level reference model that jointly encodes global scene context, objects, and inter-object relations using disentangled encoders and EMA-based knowledge distillation. We benchmark across a comprehensive suite of tasks on the SCLARO Dataset, namely zero-shot retrieval, linear probe, object detection, predicate classification, scene-graph classification, and out-of-domain generalisation. ScenarioCLIP's disentangled encoders improve over the previous works, such as PyramidCLIP's shared encoder, most notably at the object and relation levels and on out-of-domain generalisation. Code for the data generation pipeline and ScenarioCLIP is available at https://github.com/scenario-clip/SCLARO-ScenarioCLIP

Visual persuasion uses images to shape cognition, emotion, and behavior, with its effects depending on both visual attributes and semantic context. Despite recent progress, it remains unclear whether Vision-Language Models (VLMs) understand visual persuasiveness. This motivates us to ask: can VLMs assess whether an image persuasively supports an intended message, which visual factors shape this judgment, and do they align with human judgments? Through empirical analyses on image-message pairs where human raters consistently agree on the persuasiveness judgment, we show that VLMs exhibit a recall-oriented bias: they over-predict images as persuasive while achieving high recall. We introduce Visual Persuasive Factors (VPFs), a taxonomy informed by cognitive psychology for quantifying visual cues that shape persuasive judgments. Our factor-level analysis reveals that VPFs distinguish human persuasiveness judgments, whereas VLMs only partially reproduce these patterns, often generating false positives by treating persuasion-relevant cues as sufficient evidence. Building on this insight, we evaluate VPF-guided interventions and find that properly framed VPF knowledge can improve performance, but merely specifying visual cues or adding step-by-step reasoning is insufficient. By analyzing model rationales at the level of functional reasoning steps, we further identify a central bottleneck in connecting object identification to semantic message alignment.

Contrastive vision-language models continue to be the dominant approach for image-text retrieval. Contrastive Language-Image Pre-training (CLIP) trains two neural networks to align their image and text embeddings in a shared latent space. As a challenging case-study for neurosymbolic AI, recent results evaluating CLIP on negated or paraphrased text have shown mixed performance as these are difficult to define formally for text data. Negation produces the opposite meaning using various possible but small lexical changes. Paraphrasing may use very different textual expressions to denote essentially the same thing. As a result, learning of paraphrasing and negation together poses a significant challenge because of the above mismatch between changes in syntax and intended meaning expected to be captured by distances in embedding space. This paper proposes a new CLIP contrastive loss function capable of balancing the requirements of having both paraphrasing and negation. It applies training triplets consisting of original, paraphrased and negated text generated by multiple large language models to the evaluation of CLIP models. The approach, called SemCLIP, aims to learn semantically-relevant and simple embeddings, placing paraphrased captions nearer to the original image embeddings while at the same time pushing negated captions farther away. Empirically, SemCLIP is shown to be capable of preserving roughly the same performance as CLIP augmented with either negation or paraphrasing. Although direct comparisons are difficult to make because the problem of learning with both negation and paraphrasing is different, an expected benefit of SemCLIP should be robustness when applied zero-shot to downstream image classification tasks. Our experiments confirm such robustness as measured by difference in accuracy (mean-accuracy delta) between original and negated captions on five downstream datasets.

The original CLIP text encoder is limited by a maximum input length of 77 tokens, which hampers its ability to effectively process long texts and perform fine-grained semantic understanding. In addition, the CLIP text encoder lacks support for multilingual inputs. All these limitations significantly restrict its applicability across a broader range of tasks. Recent studies have attempted to replace the CLIP text encoder with an LLM-based embedder to enhance its ability in processing long texts, multilingual understanding, and fine-grained semantic comprehension. However, because the representation spaces of LLMs and the vision-language space of CLIP are pretrained independently without alignment priors, direct alignment using contrastive learning can disrupt the intrinsic vision-language alignment in the CLIP image encoder, leading to an underutilization of the knowledge acquired during pre-training. To address this challenge, we propose ProCLIP, a curriculum learning-based progressive vision-language alignment framework to effectively align the CLIP image encoder with an LLM-based embedder. Specifically, ProCLIP first distills knowledge from CLIP's text encoder into the LLM-based embedder to leverage CLIP's rich pretrained knowledge while establishing initial alignment between the LLM embedder and CLIP image encoder. Subsequently, ProCLIP further aligns the CLIP image encoder with the LLM-based embedder through image-text contrastive tuning, employing self-distillation regularization to avoid overfitting. To achieve a more effective alignment, instance semantic alignment loss and embedding structure alignment loss are employed during representation inheritance and contrastive tuning. The Code is available at https://github.com/VisionXLab/ProCLIP.

Unmanned aerial vehicles (UAVs) have become powerful platforms for real-time, high-resolution data collection, producing massive volumes of aerial videos. Efficient retrieval of relevant content from these videos is crucial for applications in urban management, emergency response, security, and disaster relief. While text-video retrieval has advanced in natural video domains, the UAV domain remains underexplored due to limitations in existing datasets, such as coarse and redundant captions. Thus, in this work, we construct the Drone Video-Text Match Dataset (DVTMD), which contains 2,864 videos and 14,320 fine-grained, semantically diverse captions. The annotations capture multiple complementary aspects, including human actions, objects, background settings, environmental conditions, and visual style, thereby enhancing text-video correspondence and reducing redundancy. Building on this dataset, we propose the Text-Conditioned Multi-granularity Alignment (TCMA) framework, which integrates global video-sentence alignment, sentence-guided frame aggregation, and word-guided patch alignment. To further refine local alignment, we design a Word and Patch Selection module that filters irrelevant content, as well as a Text-Adaptive Dynamic Temperature Mechanism that adapts attention sharpness to text type. Extensive experiments on DVTMD and CapERA establish the first complete benchmark for drone text-video retrieval. Our TCMA achieves state-of-the-art performance, including 45.5% R@1 in text-to-video and 42.8% R@1 in video-to-text retrieval, demonstrating the effectiveness of our dataset and method. The code and dataset will be released.

Adapting vision-language models in low-resource languages remains challenging, as they often produce fluent text about the wrong objects. This stems from scarce paired data, translation pivots that break alignment, and English-centric pretraining that ignores target-language semantics. We address this with a compute-aware Bengali captioning pipeline trained on LaBSE-verified EN-BN pairs and 110k bilingual-prompted synthetic images. A frozen MaxViT yields stable visual patches, a Bengali-native mBART-50 decodes, and a lightweight bridge links the modalities. Our core novelty is a tri-loss objective: Patch-Alignment Loss (PAL) aligns real and synthetic patch descriptors using decoder cross-attention, InfoNCE enforces global real-synthetic separation, and Sinkhorn-based OT ensures balanced fine-grained patch correspondence. This PAL+InfoNCE+OT synergy tightens real-synthetic patch alignment, reduces spurious matches, and drives strong gains on Flickr30k-1k (BLEU-4 12.29, METEOR 27.98, BERTScore-F1 71.20) and MSCOCO-1k (BLEU-4 12.00, METEOR 28.14, BERTScore-F1 75.40), outperforming strong CE baselines and narrowing the real-synthetic centroid gap by 41%.

Recent advances in pre-training vision-language models (VLMs), e.g., contrastive language-image pre-training (CLIP) methods, have shown great potential in learning out-of-distribution (OOD) representations. Despite showing competitive performance, the prompt-based CLIP methods still suffer from: i) inaccurate text descriptions, which leads to degraded accuracy and robustness, and poses a challenge for zero-shot CLIP methods. ii) limited vision-language embedding alignment, which is one important factor affecting generalization performance. To tackle the above issues, this paper proposes a novel Conditional Domain prompt Learning (CoDoL) method, which utilizes readily-available domain information to form prompts and contributes to improved vision-language embedding alignment, which we identify as one factor underlying the observed OOD generalization gains. To capture both instance-specific and domain-specific information, we further propose a lightweight Domain Meta Network (DMN) to generate input-conditional tokens for images in each domain. Extensive experiments on four OOD benchmarks (PACS, VLCS, OfficeHome, and DigitDG) validate the effectiveness of our proposed CoDoL method in terms of empirically improves vision-language embedding alignment across four DG benchmarks, which we present as a contributing factor (rather than the sole cause) of the observed OOD gains.

Generating semantically controllable 3D CT volumes from radiology reports requires more than a rich text encoder, it requires vision-language alignment grounded in volumetric space. Existing Text-to-CT approaches condition generation on encoders pretrained with language only or 2D vision-language objectives, providing conditioning signals that are linguistically expressive but volumetrically blind. We argue this is a structural limitation: the quality of 3D vision-language alignment, not the richness of the text encoder, is the primary bottleneck for semantic controllability in volumetric diffusion models. To address this, we propose a generation-oriented 3D-CLIP encoder trained with structured hard negatives that operate exclusively at the text level. This design increases contrastive difficulty without any additional 3D memory cost, overcoming the small-batch constraints inherent to volumetric encoders. The resulting encoder conditions a fully end-to-end latent diffusion model that operates directly in 3D latent space, eliminating the spatial artifacts and cross-slice inconsistencies introduced by super-resolution pipelines. Through systematic ablations, we establish a clear empirical link between grounding quality and downstream generative controllability. Evaluated on CT-RATE across 18 pathological conditions, our method achieves state-of-the-art performance on both image fidelity and factual correctness, while requiring less inference time and GPU memory than all competing methods. Code is at https://github.com/danielemolino/Text2CT.

It was recently observed that the representations of different models that process identical or semantically related inputs tend to align. We analyze this phenomenon using the Information Imbalance, an asymmetric rank-based measure that quantifies the capability of a representation to predict another, providing a proxy of the cross-entropy which can be computed efficiently in high-dimensional spaces. By measuring the Information Imbalance between representations generated by DeepSeek-V3 processing translations, we find that semantic information is spread across many tokens, and that semantic predictability is strongest in a set of central layers of the network, robust across six language pairs. We measure clear information asymmetries: English representations are systematically more predictive than those of other languages, and DeepSeek-V3 representations are more predictive of those in a smaller model such as Llama3-8b than the opposite. In the visual domain, we observe that semantic information concentrates in middle layers for autoregressive models and in final layers for encoder models, and these same layers yield the strongest cross-modal predictability with textual representations of image captions. Our results support the hypothesis of semantic convergence across languages, modalities, and architectures, while showing that directed predictability between representations varies strongly with layer-depth, model scale, and language.

Weakly supervised object localization (WSOL) models can predict both the object class and the spatial regions corresponding to the object, without requiring explicit bounding-box annotations. Given their reliance on classification objectives, traditional WSOL methods, like class activation mapping, tend to focus on the most discriminative object regions, often missing the full spatial extent. Although vision-language models like CLIP encode rich semantic priors, their global text and class-token embeddings are not explicitly aligned with local patch embeddings, limiting patch-level localization. Recent methods such as GenPrompt address this limitation, but at the cost of increased complexity, as they rely on conditional denoising and elaborate prompt-learning strategies. In this paper, we propose Text Distillation for Localization (TeD-Loc), which distills knowledge from CLIP text embeddings to patch embeddings through contrastive alignment, thereby enabling patch-level foreground/background localization. A localization-guided classification module is also introduced, which uses localization scores to aggregate foreground patch embeddings for joint classification and localization within a single model. In addition, a QR-based orthogonalization of class text embeddings is applied before distillation to improve discrimination for semantically similar classes. Extensive experiments show that TeD-Loc improves Top-1 Loc by ~5% on CUB and ILSVRC, and PxAP by ~31% on histopathology benchmarks, while achieving more efficient inference than GenPrompt.