Yonatan Bitton
Yonatan Bitton is a Research Scientist at Google Tel Aviv, working on vision-and-language generalization and multimodal consistency.
Research Areas
Authored Publications
Sort By
Preview abstract
Large-scale video generative models, capable of creating realistic videos of diverse visual concepts, are strong candidates for general-purpose physical world simulators. However, their adherence to physical commonsense across real-world actions remains unclear (e.g., playing tennis, backflip). Existing benchmarks suffer from limitations such as limited size, lack of human evaluation, sim-to-real gaps, and absence of fine-grained physical rule analysis. To address this, we introduce VideoPhy-2, an action-centric dataset for evaluating physical commonsense in generated videos. We curate 200 diverse actions and detailed prompts for video synthesis from modern generative models. We perform human evaluation that assesses semantic adherence, physical commonsense, and grounding of physical rules in the generated videos. Our findings reveal major shortcomings, with even the best model achieving only 22% joint performance (i.e., high semantic and physical commonsense adherence) on the hard subset of VideoPhy-2. We find that the models particularly struggle with conservation laws like mass and momentum. Finally, we also train VideoPhy-AutoEval, an automatic evaluator for fast, reliable assessment on our dataset. Overall, VideoPhy-2 serves as a rigorous benchmark, exposing critical gaps in video generative models and guiding future research in physically-grounded video generation. The data and code is available at https://videophy2.github.io/
View details
RefVNLI: Towards Scalable Evaluation of Subject-driven Text-to-image Generation
Aviv Slobodkin
Hagai Taitelbaum
Brian Gordon
Michal Sokolik
Almog Gueta
Royi Rassin
Dani Lischinski
2025
Preview abstract
Subject-driven text-to-image (T2I) generation aims to produce images that align with a given textual description, while preserving the visual identity from a referenced subject image. Despite its broad downstream applicability - ranging from enhanced personalization in image generation to consistent character representation in video rendering - progress in this field is limited by the lack of reliable automatic evaluation. Existing methods either assess only one aspect of the task (i.e., textual alignment or subject preservation), misalign with human judgments, or rely on costly API-based evaluation. To address this gap, we introduce RefVNLI, a cost-effective metric that evaluates both textual alignment and subject preservation in a single run. Trained on a large-scale dataset derived from video-reasoning benchmarks and image perturbations, RefVNLI outperforms or statistically matches existing baselines across multiple benchmarks and subject categories (e.g., Animal, Object), achieving up to 6.4-point gains in textual alignment and 5.9-point gains in subject preservation.
View details
Preview abstract
Recent advances in diffusion-based generative modeling have led to the development of text-to-video (T2V) models that can generate high-quality videos conditioned on a text prompt. Most of these T2V models often produce single-scene video clips that depict an entity performing a particular action (e.g., `a red panda climbing a tree'). However, it is pertinent to synthesize multi-scene generated videos since they are ubiquitous in the real-world (e.g., `a red panda climbing a tree' followed by `the red panda sleeps on the top of the tree'). To generate multi-scene videos from the pretrained T2V model, we introduce Time-Aligned Captions (TALC) framework. In particular, we introduce a simple and novel inductive bias to the text-conditioning mechanism in the T2V architecture that makes the model aware of the temporal alignment between the video scenes and their respective scene descriptions. As a result, we find that the T2V model can generate visually consistent (e.g., entity and background) videos that adhere to the multi-scene text descriptions. Further, we finetune the pretrained T2V model with multi-scene video-text data using the TALC framework. We observe that TALC-finetuned model outperforms the baseline methods by 15.5 points on the overall score, average of visual consistency and text adherence, across a diverse task prompts and number of generated scenes under the automatic and human evaluations.
View details
ImageInWords: Unlocking Hyper-Detailed Image Descriptions
Andrew Bunner
Ranjay Krishna
(2024)
Preview abstract
Despite the longstanding adage "an image is worth a thousand words," creating accurate and hyper-detailed image descriptions for training Vision-Language models remains challenging.
Current datasets typically have web-scraped descriptions that are short, low-granularity, and often contain details unrelated to the visual content. As a result, models trained on such data generate descriptions replete with missing information, visual inconsistencies, and hallucinations. To address these issues, we introduce ImageInWords (IIW), a carefully designed human-in-the-loop annotation framework for curating hyper-detailed image descriptions and a new dataset resulting from this process.
We validate the framework through evaluations focused on the quality of the dataset and its utility for fine-tuning with considerations for readability, comprehensiveness, specificity, hallucinations, and human-likeness. Our dataset significantly improves across these dimensions compared to recently released datasets (+66%) and GPT-4V outputs (+48%). Furthermore, models fine-tuned with IIW data excel by +31% against prior work along the same human evaluation dimensions. Given our fine-tuned models, we also evaluate text-to-image generation and vision-language reasoning. Our model's descriptions can generate images closest to the original, as judged by both automated and human metrics. We also find our model produces more compositionally rich descriptions, outperforming the best baseline by up to 6% on ARO, SVO-Probes, and Winoground datasets.
View details
DOCCI: Descriptions of Connected and Contrasting Images
Garrett Tanzer
Jaemin Cho
Su Wang
Sunayana Rane
Zack Berger
Zarana Parekh
(2024)
Preview abstract
Despite recent advancements, text-to-image (T2I) models still exhibit critical limitations, such as errors in understanding spatial relationships, object counting, text rendering, and more. One challenge in overcoming these failure modes is the lack of resources; the majority of existing image-text datasets provide only brief captions that do not offer sufficient detail to discrepancies between images and their descriptions. To advance the development of T2I models further, we introduce \textbf{Descriptions of Connected and Contrasting Images (DOCCI)}, a dataset of 15k images taken by a single person with detailed human-annotated descriptions in English. We meticulously annotated detailed and coherent descriptions, averaging 136 words, which sufficiently differentiate images from related or similar ones. We intentionally curated images that showcase a diverse range of visual properties, including entities with their attributes, various orientations, and lighting effects, many of which are related to each other. We thoroughly analyze the quality and characteristics of the image-description pairs, and assess the performance of the latest T2I and I2T models. The experimental results indicate that the current state-of-the-art T2I models still struggle with the aforementioned challenges, and even the SOTA models have not fully addressed them. DOCCI is publicly available, and we believe that this dataset will be a valuable benchmark for vision-language research.
View details
Preview abstract
The performance of instruction-following Large Vision- Language Models heavily relies on the size and quality of the instruction-tuning dataset. Existing video instruction tuning datasets are derived by prompting large language models with video captions to generate question-answer pairs, and suffer from data quality and scaling issues. While many existing video datasets have diverse labels and supervision for various tasks, their integration into Large Vision-Language Models is non-trivial. Herein, we present Video Self-Taught Reasoners (Video-STaR), a novel approach that allows the utilization of any labeled video dataset for video instruction tuning. Video-STaR uses a Large Vision-Language Model to generate video question-answer pairs off of video content and labels and utilizes the video labels to filter these question-answer pairs by only selecting correctly answered instances for instruction tuning. The filtering effectively employs the existing video labels as weak supervision for the quality of the question-answer pairs, iteratively enhancing the model through cycles of self-training until performance plateaus. Our results demonstrate that Large Vision-Language Models tuned with Video-STaR exhibit superior robustness, showing marked improvement in VQA benchmarks and adapted downstream tasks. For instance, on Kinetics700, Video-STaR improved accuracy from 50.0 to 59.9 and on zero-shot MSVD-QA from 69.7 to 71.3.
View details
VideoPhy: Evaluating Physical Commonsense for Video Generation
Kai-Wei Chang
Tianyi Xie
Hritik Bansal
Aditya Grover
Zeshun Zong
Yizhou Sun
Michal Yarom
Chenfanfu Jiang
Johnson Lin
2024
Preview abstract
Recent advances in internet-scale video data pretraining have led to the development of text-to-video generative models that can create high-quality videos across a broad range of visual concepts, synthesize realistic motions and render complex objects. Hence, these generative models have the potential to become general-purpose simulators of the physical world. However, it is unclear how far we are from this goal with the existing text-to-video generative models. To this end, we present VideoPhy, a benchmark designed to assess whether the generated videos follow physical commonsense for real-world activities (e.g. marbles will roll down when placed on a slanted surface). Specifically, we curate diverse prompts that involve interactions between various material types in the physical world (e.g., solid-solid, solid-fluid, fluid-fluid). We then generate videos conditioned on these captions from diverse state-of-the-art text-to-video generative models, including open models (e.g., CogVideoX) and closed models (e.g., Lumiere, Dream Machine). Our human evaluation reveals that the existing models severely lack the ability to generate videos adhering to the given text prompts, while also lack physical commonsense. Specifically, the best performing model, CogVideoX-5B, generates videos that adhere to the caption and physical laws for 39.6% of the instances. VideoPhy thus highlights that the video generative models are far from accurately simulating the physical world. Finally, we propose an auto-evaluator, VideoCon-Physics, to assess the performance reliably for the newly released models.
View details
A Chain-of-Thought Is as Strong as Its Weakest Link: A Benchmark for Verifiers of Reasoning Chains
Alon Jacovi
Or Honovich
Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (2024), pp. 4615–4634
Preview abstract
Prompting language models to provide step-by-step answers (e.g., “Chain-of-Thought”) is the prominent approach for complex reasoning tasks, where more accurate reasoning chains typically improve downstream task performance. Recent literature discusses automatic methods to verify reasoning to evaluate and improve their correctness. However, no fine-grained step-level datasets are available to enable thorough evaluation of such verification methods, hindering progress in this direction. We introduce REVEAL: Reasoning Verification Evaluation, a dataset to benchmark automatic verifiers of complex Chain-of-Thought reasoning in open-domain question-answering settings. REVEAL includes comprehensive labels for the relevance, attribution to evidence passages, and logical correctness of each reasoning step in a language model’s answer, across a variety of datasets and state-of-the-art language models. Evaluation on REVEAL shows that verifiers struggle at verifying reasoning chains — in particular, verifying logical correctness and detecting contradictions. Available at https://reveal-dataset.github.io/.
View details
q2d: Automatic Dialog Generation to Improve Models' Query Generation
Enav Weinreb
Ido Hakimi
Shlomi Cohen-Ganor
Yoad Lewenberg
EMNLP 2023 (2023)
Preview abstract
We propose q2d: an automatic data generation pipeline that generates information-seeking dialogues based on questions. We apply our method to create conversational versions of questions answering datasets, which we release as a new dataset. We use this data to improve query generation models, which communicate with an external search APIs to generate factual responses. Unlike previous approaches, which relied on human annotators, our method allows to automatically generate labeled dialogues with better control and scale.
In experiments, we demonstrate that: (1) Models trained on our synthetic data produce results comparable to those trained on natural data; (2) Our generated datasets are effective as a benchmark and as a training signal that generalizes to human-annotated test sets.
We also provide an extensive analysis of the quality and factuality of the generated datasets. Our studies indicate that our automatic dialogue generation pipeline is effective at improving query generation and factuality.
View details
Preview abstract
Text to image generation methods (T2I) are widely popular in generating art and other creative artifacts.
While hallucination can be a positive factor in scenarios where creativity is appreciated, such artifacts are poorly suited for tasks where the generated image needs to be grounded in a strict manner, e.g. as an illustration of a task, an action or in the context of a story.
In this paper, we propose to strengthen the factual consistency properties of T2I methods in the presence of natural prompts.
First, we cast the problem as an MT problem that translates natural prompts into visual prompts. Then we filter the image with a VQA approach where we answer a set of questions in the visual domain (the image) and in the natural language domain (the natural prompt).
Finally, to measure the alignment of answers, we depart from the recent literature that do string matching, and compare answers in an embedding space that assesses the semantic and entailment associations between a natural prompt and its generated image.
View details
