Erkut Erdem

Aarohi Srivastava, Abhinav Rastogi, Abhishek Rao et al. · allen-ai, amazon-science

Language models demonstrate both quantitative improvement and new qualitative capabilities with increasing scale. Despite their potentially transformative impact, these new capabilities are as yet poorly characterized. In order to inform future research, prepare for disruptive new model capabilities, and ameliorate socially harmful effects, it is vital that we understand the present and near-future capabilities and limitations of language models. To address this challenge, we introduce the Beyond the Imitation Game benchmark (BIG-bench). BIG-bench currently consists of 204 tasks, contributed by 450 authors across 132 institutions. Task topics are diverse, drawing problems from linguistics, childhood development, math, common-sense reasoning, biology, physics, social bias, software development, and beyond. BIG-bench focuses on tasks that are believed to be beyond the capabilities of current language models. We evaluate the behavior of OpenAI's GPT models, Google-internal dense transformer architectures, and Switch-style sparse transformers on BIG-bench, across model sizes spanning millions to hundreds of billions of parameters. In addition, a team of human expert raters performed all tasks in order to provide a strong baseline. Findings include: model performance and calibration both improve with scale, but are poor in absolute terms (and when compared with rater performance); performance is remarkably similar across model classes, though with benefits from sparsity; tasks that improve gradually and predictably commonly involve a large knowledge or memorization component, whereas tasks that exhibit "breakthrough" behavior at a critical scale often involve multiple steps or components, or brittle metrics; social bias typically increases with scale in settings with ambiguous context, but this can be improved with prompting.

Mert Cokelek, Halit Ozsoy, Nevrez Imamoglu et al.

Omnidirectional videos (ODVs) are redefining viewer experiences in virtual reality (VR) by offering an unprecedented full field-of-view (FOV). This study extends the domain of saliency prediction to 360-degree environments, addressing the complexities of spherical distortion and the integration of spatial audio. Contextually, ODVs have transformed user experience by adding a spatial audio dimension that aligns sound direction with the viewer's perspective in spherical scenes. Motivated by the lack of comprehensive datasets for 360-degree audio-visual saliency prediction, our study curates YT360-EyeTracking, a new dataset of 81 ODVs, each observed under varying audio-visual conditions. Our goal is to explore how to utilize audio-visual cues to effectively predict visual saliency in 360-degree videos. Towards this aim, we propose two novel saliency prediction models: SalViT360, a vision-transformer-based framework for ODVs equipped with spherical geometry-aware spatio-temporal attention layers, and SalViT360-AV, which further incorporates transformer adapters conditioned on audio input. Our results on a number of benchmark datasets, including our YT360-EyeTracking, demonstrate that SalViT360 and SalViT360-AV significantly outperform existing methods in predicting viewer attention in 360-degree scenes. Interpreting these results, we suggest that integrating spatial audio cues in the model architecture is crucial for accurate saliency prediction in omnidirectional videos. Code and dataset will be available at https://cyberiada.github.io/SalViT360.

Burak Ercan, Onur Eker, Aykut Erdem et al.

Event cameras are a new type of vision sensor that incorporates asynchronous and independent pixels, offering advantages over traditional frame-based cameras such as high dynamic range and minimal motion blur. However, their output is not easily understandable by humans, making the reconstruction of intensity images from event streams a fundamental task in event-based vision. While recent deep learning-based methods have shown promise in video reconstruction from events, this problem is not completely solved yet. To facilitate comparison between different approaches, standardized evaluation protocols and diverse test datasets are essential. This paper proposes a unified evaluation methodology and introduces an open-source framework called EVREAL to comprehensively benchmark and analyze various event-based video reconstruction methods from the literature. Using EVREAL, we give a detailed analysis of the state-of-the-art methods for event-based video reconstruction, and provide valuable insights into the performance of these methods under varying settings, challenging scenarios, and downstream tasks.

Nafiseh Jabbari Tofighi, Mohamed Hedi Elfkir, Nevrez Imamoglu et al.

Omnidirectional images, aka 360 images, can deliver immersive and interactive visual experiences. As their popularity has increased dramatically in recent years, evaluating the quality of 360 images has become a problem of interest since it provides insights for capturing, transmitting, and consuming this new media. However, directly adapting quality assessment methods proposed for standard natural images for omnidirectional data poses certain challenges. These models need to deal with very high-resolution data and implicit distortions due to the spherical form of the images. In this study, we present a method for no-reference 360 image quality assessment. Our proposed ST360IQ model extracts tangent viewports from the salient parts of the input omnidirectional image and employs a vision-transformers based module processing saliency selective patches/tokens that estimates a quality score from each viewport. Then, it aggregates these scores to give a final quality score. Our experiments on two benchmark datasets, namely OIQA and CVIQ datasets, demonstrate that as compared to the state-of-the-art, our approach predicts the quality of an omnidirectional image correlated with the human-perceived image quality. The code has been available on https://github.com/Nafiseh-Tofighi/ST360IQ

Osman Batur İnce, Tanin Zeraati, Semih Yagcioglu et al.

Neural networks have revolutionized language modeling and excelled in various downstream tasks. However, the extent to which these models achieve compositional generalization comparable to human cognitive abilities remains a topic of debate. While existing approaches in the field have mainly focused on novel architectures and alternative learning paradigms, we introduce a pioneering method harnessing the power of dataset cartography (Swayamdipta et al., 2020). By strategically identifying a subset of compositional generalization data using this approach, we achieve a remarkable improvement in model accuracy, yielding enhancements of up to 10% on CFQ and COGS datasets. Notably, our technique incorporates dataset cartography as a curriculum learning criterion, eliminating the need for hyperparameter tuning while consistently achieving superior performance. Our findings highlight the untapped potential of dataset cartography in unleashing the full capabilities of compositional generalization within Transformer models. Our code is available at https://github.com/cyberiada/cartography-for-compositionality.

Muhammed Burak Kizil, Enes Sanli, Niloy J. Mitra et al.

Generative video models have achieved remarkable visual fidelity and temporal coherence, yet intentional camera control remains elusive. Existing frameworks treat camera motion as a byproduct of pixel synthesis, producing trajectories that are stochastic, spatially inconsistent, and indifferent to the human subject driving the scene. In this work, we present Auteur, a method for language-driven, human-centric camera framing in generative video. Our core insight is that professional filmmakers conceive shots not as world-space trajectories but as framings defined relative to the actor, encoding shot size, angle, and composition as functions of human pose and motion. We formalize this intuition as a human-centric camera parameterization and introduce a Domain-Specific Language (DSL) that is convertible to standard 6-DoF camera parameters. A fine-tuned multimodal large language model then acts as a virtual director, mapping natural language descriptions and coarse human motion to sparse DSL keyframes that are deterministically interpolated into continuous camera trajectories, which are then provided as input to video generators. We train and evaluate Auteur on a new dataset of 34K aligned text, human motion, and DSL-annotated camera trajectories drawn from procedural synthesis and real-world movie footage from the CondensedMovies dataset. Auteur enables cinematographic framing of human-centered scenes, a capability largely absent in prior generative models. To assess this behavior, we propose new framing-focused metrics, and our experiments show that Auteur consistently outperforms existing methods

Mustafa Dogan, Ilker Kesen, Iacer Calixto et al.

As multimodal large language models (MLLMs) advance in handling interleaved image-text data, assessing their few-shot learning capabilities remains an open challenge. In this paper, we introduce FewMMBench, a comprehensive benchmark designed to evaluate MLLMs under few-shot conditions, with a focus on In-Context Learning (ICL) and Chain-of-Thought (CoT) prompting. Covering a diverse suite of multimodal understanding tasks, from attribute recognition to temporal reasoning, FewMMBench enables systematic analysis across task types, model families, and prompting strategies. We evaluate 26 open-weight MLLMs from six model families across zero-shot, few-shot, and CoT-augmented few-shot settings. Our findings reveal that instruction-tuned models exhibit strong zero-shot performance but benefit minimally, or even regress, with additional demonstrations or CoT reasoning. Retrieval-based demonstrations and increased context size also yield limited gains. These results highlight FewMMBench as a rigorous testbed for diagnosing and advancing few-shot capabilities in multimodal LLMs. The data is available at: https://huggingface.co/datasets/mustafaa/FewMMBench

Ahmet Burak Yildirim, Vedat Baday, Erkut Erdem et al.

Image inpainting task refers to erasing unwanted pixels from images and filling them in a semantically consistent and realistic way. Traditionally, the pixels that are wished to be erased are defined with binary masks. From the application point of view, a user needs to generate the masks for the objects they would like to remove which can be time-consuming and prone to errors. In this work, we are interested in an image inpainting algorithm that estimates which object to be removed based on natural language input and removes it, simultaneously. For this purpose, first, we construct a dataset named GQA-Inpaint for this task. Second, we present a novel inpainting framework, Inst-Inpaint, that can remove objects from images based on the instructions given as text prompts. We set various GAN and diffusion-based baselines and run experiments on synthetic and real image datasets. We compare methods with different evaluation metrics that measure the quality and accuracy of the models and show significant quantitative and qualitative improvements.

Ahmet Canberk Baykal, Abdul Basit Anees, Duygu Ceylan et al.

Researchers have recently begun exploring the use of StyleGAN-based models for real image editing. One particularly interesting application is using natural language descriptions to guide the editing process. Existing approaches for editing images using language either resort to instance-level latent code optimization or map predefined text prompts to some editing directions in the latent space. However, these approaches have inherent limitations. The former is not very efficient, while the latter often struggles to effectively handle multi-attribute changes. To address these weaknesses, we present CLIPInverter, a new text-driven image editing approach that is able to efficiently and reliably perform multi-attribute changes. The core of our method is the use of novel, lightweight text-conditioned adapter layers integrated into pretrained GAN-inversion networks. We demonstrate that by conditioning the initial inversion step on the CLIP embedding of the target description, we are able to obtain more successful edit directions. Additionally, we use a CLIP-guided refinement step to make corrections in the resulting residual latent codes, which further improves the alignment with the text prompt. Our method outperforms competing approaches in terms of manipulation accuracy and photo-realism on various domains including human faces, cats, and birds, as shown by our qualitative and quantitative results.

Orhan Torun, Seniha Esen Yuksel, Erkut Erdem et al.

Compared to natural images, hyperspectral images (HSIs) consist of a large number of bands, with each band capturing different spectral information from a certain wavelength, even some beyond the visible spectrum. These characteristics of HSIs make them highly effective for remote sensing applications. That said, the existing hyperspectral imaging devices introduce severe degradation in HSIs. Hence, hyperspectral image denoising has attracted lots of attention by the community lately. While recent deep HSI denoising methods have provided effective solutions, their performance under real-life complex noise remains suboptimal, as they lack adaptability to new data. To overcome these limitations, in our work, we introduce a self-modulating convolutional neural network which we refer to as SM-CNN, which utilizes correlated spectral and spatial information. At the core of the model lies a novel block, which we call spectral self-modulating residual block (SSMRB), that allows the network to transform the features in an adaptive manner based on the adjacent spectral data, enhancing the network's ability to handle complex noise. In particular, the introduction of SSMRB transforms our denoising network into a dynamic network that adapts its predicted features while denoising every input HSI with respect to its spatio-spectral characteristics. Experimental analysis on both synthetic and real data shows that the proposed SM-CNN outperforms other state-of-the-art HSI denoising methods both quantitatively and qualitatively on public benchmark datasets.

Cansu Korkmaz, A. Murat Tekalp, Zafer Dogan et al.

Flow-based generative super-resolution (SR) models learn to produce a diverse set of feasible SR solutions, called the SR space. Diversity of SR solutions increases with the temperature ($τ$) of latent variables, which introduces random variations of texture among sample solutions, resulting in visual artifacts and low fidelity. In this paper, we present a simple but effective image ensembling/fusion approach to obtain a single SR image eliminating random artifacts and improving fidelity without significantly compromising perceptual quality. We achieve this by benefiting from a diverse set of feasible photo-realistic solutions in the SR space spanned by flow models. We propose different image ensembling and fusion strategies which offer multiple paths to move sample solutions in the SR space to more desired destinations in the perception-distortion plane in a controllable manner depending on the fidelity vs. perceptual quality requirements of the task at hand. Experimental results demonstrate that our image ensembling/fusion strategy achieves more promising perception-distortion trade-off compared to sample SR images produced by flow models and adversarially trained models in terms of both quantitative metrics and visual quality.

Ilker Kesen, Andrea Pedrotti, Mustafa Dogan et al.

With the ever-increasing popularity of pretrained Video-Language Models (VidLMs), there is a pressing need to develop robust evaluation methodologies that delve deeper into their visio-linguistic capabilities. To address this challenge, we present ViLMA (Video Language Model Assessment), a task-agnostic benchmark that places the assessment of fine-grained capabilities of these models on a firm footing. Task-based evaluations, while valuable, fail to capture the complexities and specific temporal aspects of moving images that VidLMs need to process. Through carefully curated counterfactuals, ViLMA offers a controlled evaluation suite that sheds light on the true potential of these models, as well as their performance gaps compared to human-level understanding. ViLMA also includes proficiency tests, which assess basic capabilities deemed essential to solving the main counterfactual tests. We show that current VidLMs' grounding abilities are no better than those of vision-language models which use static images. This is especially striking once the performance on proficiency tests is factored in. Our benchmark serves as a catalyst for future research on VidLMs, helping to highlight areas that still need to be explored.

Mert Cokelek, Nevrez Imamoglu, Cagri Ozcinar et al.

The growing interest in omnidirectional videos (ODVs) that capture the full field-of-view (FOV) has gained 360-degree saliency prediction importance in computer vision. However, predicting where humans look in 360-degree scenes presents unique challenges, including spherical distortion, high resolution, and limited labelled data. We propose a novel vision-transformer-based model for omnidirectional videos named SalViT360 that leverages tangent image representations. We introduce a spherical geometry-aware spatiotemporal self-attention mechanism that is capable of effective omnidirectional video understanding. Furthermore, we present a consistency-based unsupervised regularization term for projection-based 360-degree dense-prediction models to reduce artefacts in the predictions that occur after inverse projection. Our approach is the first to employ tangent images for omnidirectional saliency prediction, and our experimental results on three ODV saliency datasets demonstrate its effectiveness compared to the state-of-the-art.

İlker Kesen, Aykut Erdem, Erkut Erdem et al.

This paper describes our two-stage system for the Euphemism Detection shared task hosted by the 3rd Workshop on Figurative Language Processing in conjunction with EMNLP 2022. Euphemisms tone down expressions about sensitive or unpleasant issues like addiction and death. The ambiguous nature of euphemistic words or expressions makes it challenging to detect their actual meaning within a context. In the first stage, we seek to mitigate this ambiguity by incorporating literal descriptions into input text prompts to our baseline model. It turns out that this kind of direct supervision yields remarkable performance improvement. In the second stage, we integrate visual supervision into our system using visual imageries, two sets of images generated by a text-to-image model by taking terms and descriptions as input. Our experiments demonstrate that visual supervision also gives a statistically significant performance boost. Our system achieved the second place with an F1 score of 87.2%, only about 0.9% worse than the best submission.

Moayed Haji Ali, Andrew Bond, Tolga Birdal et al.

We propose $\textbf{VidStyleODE}$, a spatiotemporally continuous disentangled $\textbf{Vid}$eo representation based upon $\textbf{Style}$GAN and Neural-$\textbf{ODE}$s. Effective traversal of the latent space learned by Generative Adversarial Networks (GANs) has been the basis for recent breakthroughs in image editing. However, the applicability of such advancements to the video domain has been hindered by the difficulty of representing and controlling videos in the latent space of GANs. In particular, videos are composed of content (i.e., appearance) and complex motion components that require a special mechanism to disentangle and control. To achieve this, VidStyleODE encodes the video content in a pre-trained StyleGAN $\mathcal{W}_+$ space and benefits from a latent ODE component to summarize the spatiotemporal dynamics of the input video. Our novel continuous video generation process then combines the two to generate high-quality and temporally consistent videos with varying frame rates. We show that our proposed method enables a variety of applications on real videos: text-guided appearance manipulation, motion manipulation, image animation, and video interpolation and extrapolation. Project website: https://cyberiada.github.io/VidStyleODE

Hatice Merve Vural, Doga Kukul, Ege Erdem Ozlu et al.

Humor is one of the few cognitive tasks where getting the reasoning right matters as much as getting the answer right. While recent work evaluates humor understanding on benchmarks such as the New Yorker Cartoon Caption Contest (NYCC), it largely treats it as black-box prediction, overlooking the structured reasoning processes underlying humor comprehension. We introduce IRS (Incongruity-Resolution Supervision), a framework that decomposes humor understanding into three components: incongruity modeling, which identifies mismatches in the visual scene; resolution modeling, which constructs coherent reinterpretations of these mismatches; and preference alignment, which evaluates candidate interpretations under human judgments. Grounded in incongruity-resolution theory and expert captionist practice, IRS supervises intermediate reasoning process through structured traces that make the path from visual perception to humorous interpretation explicit and learnable. Across 7B, 32B, and 72B models on NYCC, IRS outperforms strong open and closed multimodal baselines across caption matching and ranking tasks, with our largest model approaching expert-level performance on ranking. Zero-shot transfer to external benchmarks shows that IRS learns generalizable reasoning patterns. Our results suggest that supervising reasoning structure, rather than scale alone, is key for reasoning-centric tasks.

Mustafa Dogan, Ilker Kesen, Iacer Calixto et al.

The linguistic capabilities of Multimodal Large Language Models (MLLMs) are critical for their effective application across diverse tasks. This study aims to evaluate the performance of MLLMs on the VALSE benchmark, focusing on the efficacy of few-shot In-Context Learning (ICL), and Chain-of-Thought (CoT) prompting. We conducted a comprehensive assessment of state-of-the-art MLLMs, varying in model size and pretraining datasets. The experimental results reveal that ICL and CoT prompting significantly boost model performance, particularly in tasks requiring complex reasoning and contextual understanding. Models pretrained on captioning datasets show superior zero-shot performance, while those trained on interleaved image-text data benefit from few-shot learning. Our findings provide valuable insights into optimizing MLLMs for better grounding of language in visual contexts, highlighting the importance of the composition of pretraining data and the potential of few-shot learning strategies to improve the reasoning abilities of MLLMs.

Levent Karacan, Tolga Kerimoğlu, İsmail İnan et al.

Giving machines the ability to imagine possible new objects or scenes from linguistic descriptions and produce their realistic renderings is arguably one of the most challenging problems in computer vision. Recent advances in deep generative models have led to new approaches that give promising results towards this goal. In this paper, we introduce a new method called DiCoMoGAN for manipulating videos with natural language, aiming to perform local and semantic edits on a video clip to alter the appearances of an object of interest. Our GAN architecture allows for better utilization of multiple observations by disentangling content and motion to enable controllable semantic edits. To this end, we introduce two tightly coupled networks: (i) a representation network for constructing a concise understanding of motion dynamics and temporally invariant content, and (ii) a translation network that exploits the extracted latent content representation to actuate the manipulation according to the target description. Our qualitative and quantitative evaluations demonstrate that DiCoMoGAN significantly outperforms existing frame-based methods, producing temporally coherent and semantically more meaningful results.

Muhammed Burak Kizil, Enes Sanli, Niloy J. Mitra et al.

Video generation has achieved remarkable progress in visual fidelity and controllability, enabling conditioning on text, layout, or motion. Among these, motion control - specifying object dynamics and camera trajectories - is essential for composing complex, cinematic scenes, yet existing interfaces remain limited. We introduce LAMP that leverages large language models (LLMs) as motion planners to translate natural language descriptions into explicit 3D trajectories for dynamic objects and (relatively defined) cameras. LAMP defines a motion domain-specific language (DSL), inspired by cinematography conventions. By harnessing program synthesis capabilities of LLMs, LAMP generates structured motion programs from natural language, which are deterministically mapped to 3D trajectories. We construct a large-scale procedural dataset pairing natural text descriptions with corresponding motion programs and 3D trajectories. Experiments demonstrate LAMP's improved performance in motion controllability and alignment with user intent compared to state-of-the-art alternatives establishing the first framework for generating both object and camera motions directly from natural language specifications.

Emre Can Acikgoz, Osman Batur İnce, Rayene Bench et al.

The integration of Large Language Models (LLMs) into healthcare promises to transform medical diagnostics, research, and patient care. Yet, the progression of medical LLMs faces obstacles such as complex training requirements, rigorous evaluation demands, and the dominance of proprietary models that restrict academic exploration. Transparent, comprehensive access to LLM resources is essential for advancing the field, fostering reproducibility, and encouraging innovation in healthcare AI. We present Hippocrates, an open-source LLM framework specifically developed for the medical domain. In stark contrast to previous efforts, it offers unrestricted access to its training datasets, codebase, checkpoints, and evaluation protocols. This open approach is designed to stimulate collaborative research, allowing the community to build upon, refine, and rigorously evaluate medical LLMs within a transparent ecosystem. Also, we introduce Hippo, a family of 7B models tailored for the medical domain, fine-tuned from Mistral and LLaMA2 through continual pre-training, instruction tuning, and reinforcement learning from human and AI feedback. Our models outperform existing open medical LLMs models by a large-margin, even surpassing models with 70B parameters. Through Hippocrates, we aspire to unlock the full potential of LLMs not just to advance medical knowledge and patient care but also to democratize the benefits of AI research in healthcare, making them available across the globe.

İlker Kesen, Ozan Arkan Can, Erkut Erdem et al.

How to best integrate linguistic and perceptual processing in multi-modal tasks that involve language and vision is an important open problem. In this work, we argue that the common practice of using language in a top-down manner, to direct visual attention over high-level visual features, may not be optimal. We hypothesize that the use of language to also condition the bottom-up processing from pixels to high-level features can provide benefits to the overall performance. To support our claim, we propose a U-Net-based model and perform experiments on two language-vision dense-prediction tasks: referring expression segmentation and language-guided image colorization. We compare results where either one or both of the top-down and bottom-up visual branches are conditioned on language. Our experiments reveal that using language to control the filters for bottom-up visual processing in addition to top-down attention leads to better results on both tasks and achieves competitive performance. Our linguistic analysis suggests that bottom-up conditioning improves segmentation of objects especially when input text refers to low-level visual concepts. Code is available at https://github.com/ilkerkesen/bvpr.

Andrew Bond, Ilkin Umut Melanlioglu, Erkut Erdem et al.

Modern visual world modeling systems increasingly rely on high-capacity architectures and large-scale data to produce plausible motion, yet they often fail to preserve underlying 3D geometry or physically consistent camera dynamics. A key limitation lies not only in model capacity, but in the latent representations used to encode geometric structure. We propose S$^2$VAE, a geometry-first latent learning framework that focuses on compressing and representing the latent 3D state of a scene, including camera motion, depth, and point-level structure, rather than modeling appearance alone. Building on representations from a Visual Geometry Grounded Transformer (VGGT), we introduce a novel type of variational autoencoder using a product of Power Spherical latent distributions, explicitly enforcing hyperspherical structure in the bottleneck to preserve directional and geometric semantics under strong compression. Across depth estimation, camera pose recovery, and point cloud reconstruction, we show that geometry-aligned hyperspherical latents consistently outperform conventional Gaussian bottlenecks, particularly in high-compression regimes. Our results highlight latent geometry as a first-class design choice for physically grounded visual and world models.

Burak Can Biner, Farrin Marouf Sofian, Umur Berkay Karakaş et al.

We are witnessing a revolution in conditional image synthesis with the recent success of large scale text-to-image generation methods. This success also opens up new opportunities in controlling the generation and editing process using multi-modal input. While spatial control using cues such as depth, sketch, and other images has attracted a lot of research, we argue that another equally effective modality is audio since sound and sight are two main components of human perception. Hence, we propose a method to enable audio-conditioning in large scale image diffusion models. Our method first maps features obtained from audio clips to tokens that can be injected into the diffusion model in a fashion similar to text tokens. We introduce additional audio-image cross attention layers which we finetune while freezing the weights of the original layers of the diffusion model. In addition to audio conditioned image generation, our method can also be utilized in conjuction with diffusion based editing methods to enable audio conditioned image editing. We demonstrate our method on a wide range of audio and image datasets. We perform extensive comparisons with recent methods and show favorable performance.

Andrew Bond, Jui-Hsien Wang, Long Mai et al.

Efficient neural representations for dynamic video scenes are critical for applications ranging from video compression to interactive simulations. Yet, existing methods often face challenges related to high memory usage, lengthy training times, and temporal consistency. To address these issues, we introduce a novel neural video representation that combines 3D Gaussian splatting with continuous camera motion modeling. By leveraging Neural ODEs, our approach learns smooth camera trajectories while maintaining an explicit 3D scene representation through Gaussians. Additionally, we introduce a spatiotemporal hierarchical learning strategy, progressively refining spatial and temporal features to enhance reconstruction quality and accelerate convergence. This memory-efficient approach achieves high-quality rendering at impressive speeds. Experimental results show that our hierarchical learning, combined with robust camera motion modeling, captures complex dynamic scenes with strong temporal consistency, achieving state-of-the-art performance across diverse video datasets in both high- and low-motion scenarios.

Burak Ercan, Onur Eker, Aykut Erdem et al.

Low-light environments pose significant challenges for image enhancement methods. To address these challenges, in this work, we introduce the HUE dataset, a comprehensive collection of high-resolution event and frame sequences captured in diverse and challenging low-light conditions. Our dataset includes 106 sequences, encompassing indoor, cityscape, twilight, night, driving, and controlled scenarios, each carefully recorded to address various illumination levels and dynamic ranges. Utilizing a hybrid RGB and event camera setup. we collect a dataset that combines high-resolution event data with complementary frame data. We employ both qualitative and quantitative evaluations using no-reference metrics to assess state-of-the-art low-light enhancement and event-based image reconstruction methods. Additionally, we evaluate these methods on a downstream object detection task. Our findings reveal that while event-based methods perform well in specific metrics, they may produce false positives in practical applications. This dataset and our comprehensive analysis provide valuable insights for future research in low-light vision and hybrid camera systems.

Camila Zurdo Tagliabue, Heloisa Oss Boll, Aykut Erdem et al.

Large language models (LLMs) are increasingly used in clinical decision support, yet current evaluation methods often fail to distinguish genuine medical reasoning from superficial patterns. We introduce DeVisE (Demographics and Vital signs Evaluation), a behavioral testing framework for probing fine-grained clinical understanding. We construct a dataset of ICU discharge notes from MIMIC-IV, generating both raw (real-world) and template-based (synthetic) versions with controlled single-variable counterfactuals targeting demographic (age, gender, ethnicity) and vital sign attributes. We evaluate five LLMs spanning general-purpose and medically fine-tuned variants, under both zero-shot and fine-tuned settings. We assess model behavior via (1) input-level sensitivity - how counterfactuals alter the likelihood of a note; and (2) downstream reasoning - how they affect predicted hospital length-of-stay. Our results show that zero-shot models exhibit more coherent counterfactual reasoning patterns, while fine-tuned models tend to be more stable yet less responsive to clinically meaningful changes. Notably, demographic factors subtly but consistently influence outputs, emphasizing the importance of fairness-aware evaluation. This work highlights the utility of behavioral testing in exposing the reasoning strategies of clinical LLMs and informing the design of safer, more transparent medical AI systems.

Abdul Basit Anees, Ahmet Canberk Baykal, Muhammed Burak Kizil et al.

Generative Adversarial Networks (GANs), particularly StyleGAN and its variants, have demonstrated remarkable capabilities in generating highly realistic images. Despite their success, adapting these models to diverse tasks such as domain adaptation, reference-guided synthesis, and text-guided manipulation with limited training data remains challenging. Towards this end, in this study, we present a novel framework that significantly extends the capabilities of a pre-trained StyleGAN by integrating CLIP space via hypernetworks. This integration allows dynamic adaptation of StyleGAN to new domains defined by reference images or textual descriptions. Additionally, we introduce a CLIP-guided discriminator that enhances the alignment between generated images and target domains, ensuring superior image quality. Our approach demonstrates unprecedented flexibility, enabling text-guided image manipulation without the need for text-specific training data and facilitating seamless style transfer. Comprehensive qualitative and quantitative evaluations confirm the robustness and superior performance of our framework compared to existing methods.

Semih Yagcioglu, Osman Batur İnce, Aykut Erdem et al.

The rise of large-scale multimodal models has paved the pathway for groundbreaking advances in generative modeling and reasoning, unlocking transformative applications in a variety of complex tasks. However, a pressing question that remains is their genuine capability for stronger forms of generalization, which has been largely underexplored in the multimodal setting. Our study aims to address this by examining sequential compositional generalization using \textsc{CompAct} (\underline{Comp}ositional \underline{Act}ivities)\footnote{Project Page: \url{http://cyberiada.github.io/CompAct}}, a carefully constructed, perceptually grounded dataset set within a rich backdrop of egocentric kitchen activity videos. Each instance in our dataset is represented with a combination of raw video footage, naturally occurring sound, and crowd-sourced step-by-step descriptions. More importantly, our setup ensures that the individual concepts are consistently distributed across training and evaluation sets, while their compositions are novel in the evaluation set. We conduct a comprehensive assessment of several unimodal and multimodal models. Our findings reveal that bi-modal and tri-modal models exhibit a clear edge over their text-only counterparts. This highlights the importance of multimodality while charting a trajectory for future research in this domain.

Enes Sanli, Baris Sarper Tezcan, Aykut Erdem et al.

Recent progress in text-to-video (T2V) generation has enabled the synthesis of visually compelling and temporally coherent videos from natural language. However, these models often fall short in basic physical commonsense, producing outputs that violate intuitive expectations around causality, object behavior, and tool use. Addressing this gap, we present PhysVidBench, a benchmark designed to evaluate the physical reasoning capabilities of T2V systems. The benchmark includes 383 carefully curated prompts, emphasizing tool use, material properties, and procedural interactions, and domains where physical plausibility is crucial. For each prompt, we generate videos using diverse state-of-the-art models and adopt a three-stage evaluation pipeline: (1) formulate grounded physics questions from the prompt, (2) caption the generated video with a vision-language model, and (3) task a language model to answer several physics-involved questions using only the caption. This indirect strategy circumvents common hallucination issues in direct video-based evaluation. By highlighting affordances and tool-mediated actions, areas overlooked in current T2V evaluations, PhysVidBench provides a structured, interpretable framework for assessing physical commonsense in generative video models.

Hakan Çapuk, Andrew Bond, Muhammed Burak Kızıl et al.

Recent advances in image generation have led to remarkable improvements in synthesizing perspective images. However, these models still struggle with panoramic image generation due to unique challenges, including varying levels of geometric distortion and the requirement for seamless loop-consistency. To address these issues while leveraging the strengths of the existing models, we introduce TanDiT, a method that synthesizes panoramic scenes by generating grids of tangent-plane images covering the entire 360$^\circ$ view. Unlike previous methods relying on multiple diffusion branches, TanDiT utilizes a unified diffusion model trained to produce these tangent-plane images simultaneously within a single denoising iteration. Furthermore, we propose a model-agnostic post-processing step specifically designed to enhance global coherence across the generated panoramas. To accurately assess panoramic image quality, we also present two specialized metrics, TangentIS and TangentFID, and provide a comprehensive benchmark comprising captioned panoramic datasets and standardized evaluation scripts. Extensive experiments demonstrate that our method generalizes effectively beyond its training data, robustly interprets detailed and complex text prompts, and seamlessly integrates with various generative models to yield high-quality, diverse panoramic images.

Enes Karanfil, Nevrez Imamoglu, Erkut Erdem et al.

Scene understanding in remote sensing often faces challenges in generating accurate representations for complex environments such as various land use areas or coastal regions, which may also include snow, clouds, or haze. To address this, we present a vision-language framework named Spectral LLaVA, which integrates multispectral data with vision-language alignment techniques to enhance scene representation and description. Using the BigEarthNet v2 dataset from Sentinel-2, we establish a baseline with RGB-based scene descriptions and further demonstrate substantial improvements through the incorporation of multispectral information. Our framework optimizes a lightweight linear projection layer for alignment while keeping the vision backbone of SpectralGPT frozen. Our experiments encompass scene classification using linear probing and language modeling for jointly performing scene classification and description generation. Our results highlight Spectral LLaVA's ability to produce detailed and accurate descriptions, particularly for scenarios where RGB data alone proves inadequate, while also enhancing classification performance by refining SpectralGPT features into semantically meaningful representations.

Yigit Ekin, Ahmet Burak Yildirim, Erdem Eren Caglar et al.

Advanced image editing techniques, particularly inpainting, are essential for seamlessly removing unwanted elements while preserving visual integrity. Traditional GAN-based methods have achieved notable success, but recent advancements in diffusion models have produced superior results due to their training on large-scale datasets, enabling the generation of remarkably realistic inpainted images. Despite their strengths, diffusion models often struggle with object removal tasks without explicit guidance, leading to unintended hallucinations of the removed object. To address this issue, we introduce CLIPAway, a novel approach leveraging CLIP embeddings to focus on background regions while excluding foreground elements. CLIPAway enhances inpainting accuracy and quality by identifying embeddings that prioritize the background, thus achieving seamless object removal. Unlike other methods that rely on specialized training datasets or costly manual annotations, CLIPAway provides a flexible, plug-and-play solution compatible with various diffusion-based inpainting techniques.

Burak Ercan, Onur Eker, Canberk Saglam et al.

Event-based cameras are becoming increasingly popular for their ability to capture high-speed motion with low latency and high dynamic range. However, generating videos from events remains challenging due to the highly sparse and varying nature of event data. To address this, in this study, we propose HyperE2VID, a dynamic neural network architecture for event-based video reconstruction. Our approach uses hypernetworks to generate per-pixel adaptive filters guided by a context fusion module that combines information from event voxel grids and previously reconstructed intensity images. We also employ a curriculum learning strategy to train the network more robustly. Our comprehensive experimental evaluations across various benchmark datasets reveal that HyperE2VID not only surpasses current state-of-the-art methods in terms of reconstruction quality but also achieves this with fewer parameters, reduced computational requirements, and accelerated inference times.

Adil Kaan Akan, Erkut Erdem, Aykut Erdem et al.

Motion is an important cue for video prediction and often utilized by separating video content into static and dynamic components. Most of the previous work utilizing motion is deterministic but there are stochastic methods that can model the inherent uncertainty of the future. Existing stochastic models either do not reason about motion explicitly or make limiting assumptions about the static part. In this paper, we reason about appearance and motion in the video stochastically by predicting the future based on the motion history. Explicit reasoning about motion without history already reaches the performance of current stochastic models. The motion history further improves the results by allowing to predict consistent dynamics several frames into the future. Our model performs comparably to the state-of-the-art models on the generic video prediction datasets, however, significantly outperforms them on two challenging real-world autonomous driving datasets with complex motion and dynamic background.

Aysun Kocak, Erkut Erdem, Aykut Erdem

Predicting saliency in videos is a challenging problem due to complex modeling of interactions between spatial and temporal information, especially when ever-changing, dynamic nature of videos is considered. Recently, researchers have proposed large-scale datasets and models that take advantage of deep learning as a way to understand what's important for video saliency. These approaches, however, learn to combine spatial and temporal features in a static manner and do not adapt themselves much to the changes in the video content. In this paper, we introduce Gated Fusion Network for dynamic saliency (GFSalNet), the first deep saliency model capable of making predictions in a dynamic way via gated fusion mechanism. Moreover, our model also exploits spatial and channel-wise attention within a multi-scale architecture that further allows for highly accurate predictions. We evaluate the proposed approach on a number of datasets, and our experimental analysis demonstrates that it outperforms or is highly competitive with the state of the art. Importantly, we show that it has a good generalization ability, and moreover, exploits temporal information more effectively via its adaptive fusion scheme.

Ahmet E. Tekden, Aykut Erdem, Erkut Erdem et al.

Pushing is an essential non-prehensile manipulation skill used for tasks ranging from pre-grasp manipulation to scene rearrangement, reasoning about object relations in the scene, and thus pushing actions have been widely studied in robotics. The effective use of pushing actions often requires an understanding of the dynamics of the manipulated objects and adaptation to the discrepancies between prediction and reality. For this reason, effect prediction and parameter estimation with pushing actions have been heavily investigated in the literature. However, current approaches are limited because they either model systems with a fixed number of objects or use image-based representations whose outputs are not very interpretable and quickly accumulate errors. In this paper, we propose a graph neural network based framework for effect prediction and parameter estimation of pushing actions by modeling object relations based on contacts or articulations. Our framework is validated both in real and simulated environments containing different shaped multi-part objects connected via different types of joints and objects with different masses, and it outperforms image-based representations on physics prediction. Our approach enables the robot to predict and adapt the effect of a pushing action as it observes the scene. It can also be used for tool manipulation with never-seen tools. Further, we demonstrate 6D effect prediction in the lever-up action in the context of robot-based hard-disk disassembly.

Ozan Caglayan, Menekse Kuyu, Mustafa Sercan Amac et al.

Pre-trained language models have been shown to improve performance in many natural language tasks substantially. Although the early focus of such models was single language pre-training, recent advances have resulted in cross-lingual and visual pre-training methods. In this paper, we combine these two approaches to learn visually-grounded cross-lingual representations. Specifically, we extend the translation language modelling (Lample and Conneau, 2019) with masked region classification and perform pre-training with three-way parallel vision & language corpora. We show that when fine-tuned for multimodal machine translation, these models obtain state-of-the-art performance. We also provide qualitative insights into the usefulness of the learned grounded representations.

Begum Citamak, Ozan Caglayan, Menekse Kuyu et al.

Automatic generation of video descriptions in natural language, also called video captioning, aims to understand the visual content of the video and produce a natural language sentence depicting the objects and actions in the scene. This challenging integrated vision and language problem, however, has been predominantly addressed for English. The lack of data and the linguistic properties of other languages limit the success of existing approaches for such languages. In this paper we target Turkish, a morphologically rich and agglutinative language that has very different properties compared to English. To do so, we create the first large scale video captioning dataset for this language by carefully translating the English descriptions of the videos in the MSVD (Microsoft Research Video Description Corpus) dataset into Turkish. In addition to enabling research in video captioning in Turkish, the parallel English-Turkish descriptions also enables the study of the role of video context in (multimodal) machine translation. In our experiments, we build models for both video captioning and multimodal machine translation and investigate the effect of different word segmentation approaches and different neural architectures to better address the properties of Turkish. We hope that the MSVD-Turkish dataset and the results reported in this work will lead to better video captioning and multimodal machine translation models for Turkish and other morphology rich and agglutinative languages.

Tayfun Ates, M. Samil Atesoglu, Cagatay Yigit et al.

Humans are able to perceive, understand and reason about causal events. Developing models with similar physical and causal understanding capabilities is a long-standing goal of artificial intelligence. As a step towards this direction, we introduce CRAFT, a new video question answering dataset that requires causal reasoning about physical forces and object interactions. It contains 58K video and question pairs that are generated from 10K videos from 20 different virtual environments, containing various objects in motion that interact with each other and the scene. Two question categories in CRAFT include previously studied descriptive and counterfactual questions. Additionally, inspired by the Force Dynamics Theory in cognitive linguistics, we introduce a new causal question category that involves understanding the causal interactions between objects through notions like cause, enable, and prevent. Our results show that even though the questions in CRAFT are easy for humans, the tested baseline models, including existing state-of-the-art methods, do not yet deal with the challenges posed in our benchmark.

Ahmet Serdar Karadeniz, Erkut Erdem, Aykut Erdem

Capturing images under extremely low-light conditions poses significant challenges for the standard camera pipeline. Images become too dark and too noisy, which makes traditional enhancement techniques almost impossible to apply. Recently, learning-based approaches have shown very promising results for this task since they have substantially more expressive capabilities to allow for improved quality. Motivated by these studies, in this paper, we aim to leverage burst photography to boost the performance and obtain much sharper and more accurate RGB images from extremely dark raw images. The backbone of our proposed framework is a novel coarse-to-fine network architecture that generates high-quality outputs progressively. The coarse network predicts a low-resolution, denoised raw image, which is then fed to the fine network to recover fine-scale details and realistic textures. To further reduce the noise level and improve the color accuracy, we extend this network to a permutation invariant structure so that it takes a burst of low-light images as input and merges information from multiple images at the feature-level. Our experiments demonstrate that our approach leads to perceptually more pleasing results than the state-of-the-art methods by producing more detailed and considerably higher quality images.

Ahmet Serdar Karadeniz, Erkut Erdem, Aykut Erdem

Capturing images under extremely low-light conditions poses significant challenges for the standard camera pipeline. Images become too dark and too noisy, which makes traditional image enhancement techniques almost impossible to apply. Very recently, researchers have shown promising results using learning based approaches. Motivated by these ideas, in this paper, we propose a deep learning framework for obtaining clean and colorful RGB images from extremely dark raw images. The backbone of our framework is a novel coarse-to-fine network architecture that generates high-quality outputs in a progressive manner. The coarse network predicts a low-resolution, denoised raw image, which is then fed to the fine network to recover fine-scale details and realistic textures. To further reduce noise and improve color accuracy, we extend this network to a permutation invariant structure so that it takes a burst of low-light images as input and merges information from multiple images at the feature-level. Our experiments demonstrate that the proposed approach leads to perceptually more pleasing results than state-of-the-art methods by producing much sharper and higher quality images.

Mahmut Yurt, Salman Ul Hassan Dar, Aykut Erdem et al.

Multi-contrast MRI protocols increase the level of morphological information available for diagnosis. Yet, the number and quality of contrasts is limited in practice by various factors including scan time and patient motion. Synthesis of missing or corrupted contrasts can alleviate this limitation to improve clinical utility. Common approaches for multi-contrast MRI involve either one-to-one and many-to-one synthesis methods. One-to-one methods take as input a single source contrast, and they learn a latent representation sensitive to unique features of the source. Meanwhile, many-to-one methods receive multiple distinct sources, and they learn a shared latent representation more sensitive to common features across sources. For enhanced image synthesis, here we propose a multi-stream approach that aggregates information across multiple source images via a mixture of multiple one-to-one streams and a joint many-to-one stream. The shared feature maps generated in the many-to-one stream and the complementary feature maps generated in the one-to-one streams are combined with a fusion block. The location of the fusion block is adaptively modified to maximize task-specific performance. Qualitative and quantitative assessments on T1-, T2-, PD-weighted and FLAIR images clearly demonstrate the superior performance of the proposed method compared to previous state-of-the-art one-to-one and many-to-one methods.

Mustafa Sercan Amac, Semih Yagcioglu, Aykut Erdem et al.

This paper addresses the problem of comprehending procedural commonsense knowledge. This is a challenging task as it requires identifying key entities, keeping track of their state changes, and understanding temporal and causal relations. Contrary to most of the previous work, in this study, we do not rely on strong inductive bias and explore the question of how multimodality can be exploited to provide a complementary semantic signal. Towards this end, we introduce a new entity-aware neural comprehension model augmented with external relational memory units. Our model learns to dynamically update entity states in relation to each other while reading the text instructions. Our experimental analysis on the visual reasoning tasks in the recently proposed RecipeQA dataset reveals that our approach improves the accuracy of the previously reported models by a large margin. Moreover, we find that our model learns effective dynamic representations of entities even though we do not use any supervision at the level of entity states.

Ahmet E. Tekden, Aykut Erdem, Erkut Erdem et al.

In recent years, graph neural networks have been successfully applied for learning the dynamics of complex and partially observable physical systems. However, their use in the robotics domain is, to date, still limited. In this paper, we introduce Belief Regulated Dual Propagation Networks (BRDPN), a general-purpose learnable physics engine, which enables a robot to predict the effects of its actions in scenes containing groups of articulated multi-part objects. Specifically, our framework extends recently proposed propagation networks (PropNets) and consists of two complementary components, a physics predictor and a belief regulator. While the former predicts the future states of the object(s) manipulated by the robot, the latter constantly corrects the robot's knowledge regarding the objects and their relations. Our results showed that after training in a simulator, the robot can reliably predict the consequences of its actions in object trajectory level and exploit its own interaction experience to correct its belief about the state of the environment, enabling better predictions in partially observable environments. Furthermore, the trained model was transferred to the real world and verified in predicting trajectories of pushed interacting objects whose joint relations were initially unknown. We compared BRDPN against PropNets, and showed that BRDPN performs consistently well. Moreover, BRDPN can adapt its physic predictions, since the relations can be predicted online.

Semih Yagcioglu, Aykut Erdem, Erkut Erdem et al.

Understanding and reasoning about cooking recipes is a fruitful research direction towards enabling machines to interpret procedural text. In this work, we introduce RecipeQA, a dataset for multimodal comprehension of cooking recipes. It comprises of approximately 20K instructional recipes with multiple modalities such as titles, descriptions and aligned set of images. With over 36K automatically generated question-answer pairs, we design a set of comprehension and reasoning tasks that require joint understanding of images and text, capturing the temporal flow of events and making sense of procedural knowledge. Our preliminary results indicate that RecipeQA will serve as a challenging test bed and an ideal benchmark for evaluating machine comprehension systems. The data and leaderboard are available at http://hucvl.github.io/recipeqa.

Levent Karacan, Zeynep Akata, Aykut Erdem et al.

In this study, we explore building a two-stage framework for enabling users to directly manipulate high-level attributes of a natural scene. The key to our approach is a deep generative network which can hallucinate images of a scene as if they were taken at a different season (e.g. during winter), weather condition (e.g. in a cloudy day) or time of the day (e.g. at sunset). Once the scene is hallucinated with the given attributes, the corresponding look is then transferred to the input image while preserving the semantic details intact, giving a photo-realistic manipulation result. As the proposed framework hallucinates what the scene will look like, it does not require any reference style image as commonly utilized in most of the appearance or style transfer approaches. Moreover, it allows to simultaneously manipulate a given scene according to a diverse set of transient attributes within a single model, eliminating the need of training multiple networks per each translation task. Our comprehensive set of qualitative and quantitative results demonstrate the effectiveness of our approach against the competing methods.

Mehmet Günel, Erkut Erdem, Aykut Erdem

Developing techniques for editing an outfit image through natural sentences and accordingly generating new outfits has promising applications for art, fashion and design. However, it is considered as a certainly challenging task since image manipulation should be carried out only on the relevant parts of the image while keeping the remaining sections untouched. Moreover, this manipulation process should generate an image that is as realistic as possible. In this work, we propose FiLMedGAN, which leverages feature-wise linear modulation (FiLM) to relate and transform visual features with natural language representations without using extra spatial information. Our experiments demonstrate that this approach, when combined with skip connections and total variation regularization, produces more plausible results than the baseline work, and has a better localization capability when generating new outfits consistent with the target description.

Salman Ul Hassan Dar, Mahmut Yurt, Levent Karacan et al.

Acquiring images of the same anatomy with multiple different contrasts increases the diversity of diagnostic information available in an MR exam. Yet, scan time limitations may prohibit acquisition of certain contrasts, and images for some contrast may be corrupted by noise and artifacts. In such cases, the ability to synthesize unacquired or corrupted contrasts from remaining contrasts can improve diagnostic utility. For multi-contrast synthesis, current methods learn a nonlinear intensity transformation between the source and target images, either via nonlinear regression or deterministic neural networks. These methods can in turn suffer from loss of high-spatial-frequency information in synthesized images. Here we propose a new approach for multi-contrast MRI synthesis based on conditional generative adversarial networks. The proposed approach preserves high-frequency details via an adversarial loss; and it offers enhanced synthesis performance via a pixel-wise loss for registered multi-contrast images and a cycle-consistency loss for unregistered images. Information from neighboring cross-sections are utilized to further improved synthesis quality. Demonstrations on T1- and T2-weighted images from healthy subjects and patients clearly indicate the superior performance of the proposed approach compared to previous state-of-the-art methods. Our synthesis approach can help improve quality and versatility of multi-contrast MRI exams without the need for prolonged examinations.

Mert Kilickaya, Aykut Erdem, Nazli Ikizler-Cinbis et al.

The task of generating natural language descriptions from images has received a lot of attention in recent years. Consequently, it is becoming increasingly important to evaluate such image captioning approaches in an automatic manner. In this paper, we provide an in-depth evaluation of the existing image captioning metrics through a series of carefully designed experiments. Moreover, we explore the utilization of the recently proposed Word Mover's Distance (WMD) document metric for the purpose of image captioning. Our findings outline the differences and/or similarities between metrics and their relative robustness by means of extensive correlation, accuracy and distraction based evaluations. Our results also demonstrate that WMD provides strong advantages over other metrics.

Levent Karacan, Zeynep Akata, Aykut Erdem et al.

Automatic image synthesis research has been rapidly growing with deep networks getting more and more expressive. In the last couple of years, we have observed images of digits, indoor scenes, birds, chairs, etc. being automatically generated. The expressive power of image generators have also been enhanced by introducing several forms of conditioning variables such as object names, sentences, bounding box and key-point locations. In this work, we propose a novel deep conditional generative adversarial network architecture that takes its strength from the semantic layout and scene attributes integrated as conditioning variables. We show that our architecture is able to generate realistic outdoor scene images under different conditions, e.g. day-night, sunny-foggy, with clear object boundaries.