Recognizing and reasoning about occluded (partially or fully hidden) objects is vital to understanding visual scenes, as occlusions frequently occur in real-world environments and act as obstacles for spatial comprehension. To test models' ability to reason about multiple occluded objects, we introduce a novel task, Counting Amodally for Patterns Through Unseen REgions (CAPTURe), which requires a model to count objects arranged in a pattern by inferring how the pattern continues behind an occluder (an object which blocks parts of the scene). CAPTURe requires both recognizing visual patterns and reasoning, making it a useful testbed for evaluating vision-language models (VLMs) on whether they understand occluded patterns and possess spatial understanding skills. By requiring models to reason about occluded objects, CAPTURe also tests VLMs' ability to form world models that would allow them to fill in missing information. CAPTURe consists of two parts: (1) CAPTURe-real, with manually filtered images of real objects in patterns and (2) CAPTURe-synthetic, a controlled diagnostic with generated patterned images. We evaluate four strong VLMs (GPT-4o, Intern-VL2, Molmo, and Qwen2-VL) on CAPTURe, finding that models struggle to count on both occluded and unoccluded patterns. Crucially, we find that models perform worse with occlusion, suggesting that VLMs are also deficient in inferring unseen spatial relationships: even the strongest VLMs like GPT-4o fail to count with occlusion. In contrast, we find that humans achieve very little error on CAPTURe. We also find that providing auxiliary information of occluded object locations increases performance, underscoring that the model error comes both from an inability to handle occlusion as well as difficulty in counting in images. Code and data: https://github.com/atinpothiraj/CAPTURe

The process of creating training data to teach models is currently driven by humans, who manually analyze model weaknesses and plan how to create data that improves a student model. Recent approaches using large language models (LLMs) as annotators reduce human annotation effort, but still require humans to interpret feedback from evaluations and control the LLM to produce data the student needs. Automating this labor-intensive process by creating autonomous data generation agents – or teachers – is desirable, but requires environments that can simulate the feedback-driven, iterative, closed loop of data creation. To enable rapid and scalable testing for such agents and their modules, we introduce DataEnvGym, a testbed of teacher environments for data generation agents. DataEnvGym frames data generation as a sequential decision-making task, involving an agent consisting of a data generation policy (which generates a plan for creating training data) and a data generation engine (which transforms the plan into data), inside an environment that provides feedback from a student. The agent’s end goal is to improve student model performance. Students are iteratively trained and evaluated on generated data, with their feedback (in the form of errors or weak skills) being reported to the agent after each iteration. As a general-purpose testbed, DataEnvGym includes multiple instantiations of teacher environments across three levels of structure in the state representation and action space, with varying levels of scaffolding support. More structured environments are based on automatically-inferred skills and offer a higher degree of interpretability and control over the curriculum. We support developing and testing data generation agents in four diverse tasks covering text, images, and actions (mathematics, programming, visual question answering, and tool-use) and test multiple student and teacher models. We find that example agents in our teaching environments can iteratively improve students across diverse tasks and settings. Moreover, we show that environments can teach different skill levels and can be used to test variants of key modules, pointing to directions of future work in improving data generation agents, engines, and feedback mechanisms. Project page: https://DataEnvGym.github.io.

ControlNets are widely used for adding spatial control to text-to-image diffusion models. However, when it comes to controllable video generation, ControlNets cannot be directly integrated into new backbones due to feature space mismatches, and training ControlNets for new backbones can be a significant burden for many users. Furthermore, applying ControlNets independently to different frames can not effectively maintain object temporal consistency. To address these challenges, we introduce Ctrl-Adapter, an efficient and versatile framework that adds diverse controls to any image/video diffusion models through the adaptation of pretrained ControlNets. Ctrl-Adapter offers strong and diverse capabilities, including image and video control, sparse-frame video control, fine-grained patch-level multi-condition control, zero-shot adaptation to unseen conditions, and supports a variety of downstream tasks beyond spatial control, including video editing, video style transfer, and text-guided motion control. With six diverse U-Net/DiT-based image/video diffusion models (SDXL, PixArt-α, I2VGen-XL, SVD, Latte, Hotshot-XL), Ctrl-Adapter matches the performance of pretrained ControlNets on COCO and achieves the state-of-the-art on DAVIS 2017 with significantly lower computation (< 10 GPU hours).

There is growing interest in integrating high-fidelity visual synthesis capabilities into large language models (LLMs) without compromising their strong reasoning capabilities. Existing methods that directly train LLMs or bridge LLMs and diffusion models usually suffer from costly training since the backbone LLMs have not seen image representations during pretraining. We present Bifrost-1, a unified framework that bridges pretrained multimodal LLMs (MLLMs) and diffusion models using patch-level CLIP image embeddings as latent variables, which are natively aligned with the MLLM's CLIP visual encoder. These patch-level image embeddings are integrated into the diffusion model with a lightweight adaptation of its ControlNet. To retain the original multimodal reasoning capabilities of MLLMs, we equip the MLLM with a visual generation branch initialized from the original MLLM parameters when predicting the patch-level image embeddings. By seamlessly integrating pretrained MLLMs and diffusion models with patch-level CLIP latents, our framework enables high-fidelity controllable image generation with significant training efficiency. Our experiments demonstrate that Bifrost-1 achieves comparable or better performance than previous methods in terms of visual fidelity and multimodal understanding, with substantially lower compute during training. We also provide comprehensive ablation studies showing the effectiveness of our design choices. Project page: https://bifrost-1.github.io.

The goal of interactive image segmentation is to delineate specific regions within an image via visual or language prompts. Low-latency and high-quality interactive segmentation with diverse prompts remain challenging for existing specialist and generalist models. Specialist models, with their limited prompts and task-specific designs, experience high latency because the image must be recomputed every time the prompt is updated, due to the joint encoding of image and visual prompts. Generalist models, exemplified by the Segment Anything Model (SAM), have recently excelled in prompt diversity and efficiency, lifting image segmentation to the foundation model era. However, for high-quality segmentations, SAM still lags behind state-of-the-art specialist models despite SAM being trained with ×100 more segmentation masks. In this work, we delve deep into the architectural differences between the two types of models. We observe that dense representation and fusion of visual prompts are the key design choices contributing to the high segmentation quality of specialist models. In light of this, we reintroduce this dense design into the generalist models, to facilitate the development of generalist models with high segmentation quality. To densely represent diverse visual prompts, we propose to use a dense map to capture five types: clicks, boxes, polygons, scribbles, and masks. Thus, we propose SegNext, a next-generation interactive segmentation approach offering low latency, high quality, and diverse prompt support. Our method outperforms current state-of-the-art methods on HQSeg-44K and DAVIS, both quantitatively and qualitatively.

Evaluating text-to-image models is notoriously difficult. A strong recent approach for assessing text-image faithfulness is based on QG/A (question generation and answering), which uses pre-trained foundational models to automatically generate a set of questions and answers from the prompt, and output images are scored based on whether these answers extracted with a visual question answering model are consistent with the prompt-based answers. This kind of evaluation is naturally dependent on the quality of the underlying QG and VQA models. We identify and address several reliability challenges in existing QG/A work: (a) QG questions should respect the prompt (avoiding hallucinations, duplications, and omissions) and (b) VQA answers should be consistent (not asserting that there is no motorcycle in an image while also claiming the motorcycle is blue). We address these issues with Davidsonian Scene Graph (DSG), an empirically grounded evaluation framework inspired by formal semantics, which is adaptable to any QG/A frameworks. DSG produces atomic and unique questions organized in dependency graphs, which (i) ensure appropriate semantic coverage and (ii) sidestep inconsistent answers. With extensive experimentation and human evaluation on a range of model configurations (LLM, VQA, and T2I), we empirically demonstrate that DSG addresses the challenges noted above. Finally, we present DSG-1k, an open-sourced evaluation benchmark that includes 1,060 prompts, covering a wide range of fine-grained semantic categories with a balanced distribution. We release the DSG-1k prompts and the corresponding DSG questions.

Recent text-to-image (T2I) generation models have demonstrated impressive capabilities in creating images from text descriptions. However, these T2I generation models often fail to generate images that precisely match the details of the text inputs, such as incorrect spatial relationship or missing objects. In this paper, we introduce SELMA: Skill-Specific Expert Learning and Merging with Auto-Generated Data, a novel paradigm to improve the faithfulness of T2I models by fine-tuning models on automatically generated, multi-skill image-text datasets, with skill-specific expert learning and merging. First, SELMA leverages an LLM’s in-context learning capability to generate multiple datasets of text prompts that can teach different skills, and then generates the images with a T2I model based on the prompts. Next, SELMA adapts the T2I model to the new skills by learning multiple single-skill LoRA (low-rank adaptation) experts followed by expert merging. Our independent expert fine-tuning specializes multiple models for different skills, and expert merging helps build a joint multi-skill T2I model that can generate faithful images given diverse text prompts, while mitigating the knowledge conflict from different datasets. We empirically demonstrate that SELMA significantly improves the semantic alignment and text faithfulness of state-of-the-art T2I diffusion models on multiple benchmarks (+2.1% on TIFA and +6.9% on DSG), human preference metrics (PickScore, ImageReward, and HPS), as well as human evaluation. Moreover, fine-tuning with image-text pairs auto-collected via SELMA shows comparable performance to fine-tuning with ground truth data. Lastly, we show that fine-tuning with images from a weaker T2I model can help improve the generation quality of a stronger T2I model, suggesting promising weak-to-strong generalization in T2I models. We provide code in the supplementary materials.

There is growing interest in searching for information from large video corpora. Prior works have studied relevant tasks, such as text-based video retrieval, moment retrieval, video summarization, and video captioning in isolation, without an end-to-end setup that can jointly search from video corpora and generate summaries. Such an end-to-end setup would allow for many interesting applications, e.g., a text-based search that finds a relevant video from a video corpus, extracts the most relevant moment from that video, and segments the moment into important steps with captions. To address this, we present the HIREST (HIerarchical REtrieval and STep-captioning) dataset and propose a new benchmark that covers hierarchical information retrieval and visual/textual stepwise summarization from an instructional video corpus. Hirest consists of 3.4K text-video pairs from an instructional video dataset, where 1.1 K videos have annotations of moment spans relevant to text query and breakdown of each moment into key instruction steps with caption and timestamps (totaling 8.6K step captions). Our hierarchical benchmark consists of video retrieval, moment retrieval, and two novel moment segmentation and step captioning tasks. In moment segmentation, models break down a video moment into instruction steps and identify start-end boundaries. In step captioning, models generate a textual summary for each step. We also present starting point task-specific and end-to-end joint baseline models for our new benchmark. While the baseline models show some promising results, there still exists large room for future improvement by the community.11code and data: https://github.com/j-min/HiREST

Recently, DALL-E [45], a multimodal transformer language model, and its variants including diffusion models have shown high-quality text-to-image generation capabilities. However, despite the realistic image generation results, there has not been a detailed analysis of how to evaluate such models. In this work, we investigate the visual reasoning capabilities and social biases of different text-to-image models, covering both multimodal transformer language models and diffusion models. First, we measure three visual reasoning skills: object recognition, object counting, and spatial relation understanding. For this, we propose PaintSkills, a compositional diagnostic evaluation dataset that measures these skills. Despite the high-fidelity image generation capability, a large gap exists between the performance of recent models and the upper bound accuracy in object counting and spatial relation understanding skills. Second, we assess the gender and skin tone biases by measuring the gender/skin tone distribution of generated images across various professions and attributes. We demonstrate that recent text-to-image generation models learn specific biases about gender and skin tone from web image-text pairs. We hope our work will help guide future progress in improving text-to-image generation models on visual reasoning skills and learning socially unbiased representations.1

As large language models have demonstrated impressive performance in many domains, recent works have adopted language models (LMs) as controllers of visual modules for vision-and-language tasks. While existing work focuses on equipping LMs with visual understanding, we propose two novel interpretable/explainable visual programming frameworks for text-to-image (T2I) generation and evaluation. First, we introduce VPGen, an interpretable step-by-step T2I generation framework that decomposes T2I generation into three steps: object/count generation, layout generation, and image generation. We employ an LM to handle the first two steps (object/count generation and layout generation), by finetuning it on text-layout pairs. Our step-by-step T2I generation framework provides stronger spatial control than end-to-end models, the dominant approach for this task. Furthermore, we leverage the world knowledge of pretrained LMs, overcoming the limitation of previous layout-guided T2I works that can only handle predefined object classes. We demonstrate that our VPGen has improved control in counts/spatial relations/scales of objects than state-of-the-art T2I generation models. Second, we introduce VPEval, an interpretable and explainable evaluation framework for T2I generation based on visual programming. Unlike previous T2I evaluations with a single scoring model that is accurate in some skills but unreliable in others, VPEval produces evaluation programs that invoke a set of visual modules that are experts in different skills, and also provides visual+textual explanations of the evaluation results. Our analysis shows that VPEval provides a more human-correlated evaluation for skill-specific and open-ended prompts than widely used single model-based evaluation. We hope that our work encourages future progress on interpretable/explainable generation and evaluation for T2I models.

Action knowledge involves the understanding of textual, visual, and temporal aspects of actions. We introduce the **Action Dynamics Benchmark (ActionBench)** containing two carefully designed probing tasks: Action Antonym and Video Reversal, which targets multimodal alignment capabilities and temporal understanding skills of the model, respectively. Despite recent video-language models’ (VidLM) impressive performance on various benchmark tasks, our diagnostic tasks reveal their surprising deficiency (near-random performance) in action knowledge, suggesting that current models rely on object recognition abilities as a shortcut for action understanding. To remedy this, we propose a novel framework, **Paxion**, along with a new **Discriminative Video Dynamics Modeling (DVDM)** objective. The Paxion framework utilizes a **Knowledge Patcher** network to encode new action knowledge and a **Knowledge Fuser** component to integrate the Patcher into frozen VidLMs without compromising their existing capabilities. Due to limitations of the widely-used Video-Text Contrastive (VTC) loss for learning action knowledge, we introduce the DVDM objective to train the Knowledge Patcher. DVDM forces the model to encode the correlation between the action text and the correct ordering of video frames. Our extensive analyses show that Paxion and DVDM together effectively fill the gap in action knowledge understanding (~50% → 80%), while maintaining or improving performance on a wide spectrum of both object- and action-centric downstream tasks.

Recent studies have shown promising results on utilizing large pre-trained image-language models for video question answering. While these image-language models can efficiently bootstrap the representation learning of video-language models, they typically concatenate uniformly sampled video frames as visual inputs without explicit language-aware, temporal modeling. When only a portion of a video input is relevant to the language query, such uniform frame sampling can often lead to missing important visual cues. Although humans often find a video moment to focus on and rewind the moment to answer questions, training a query-aware video moment localizer often requires expensive annotations and high computational costs. To address this issue, we propose Self-Chained Video Localization-Answering (SeViLA), a novel framework that leverages a single image-language model (BLIP- 2) to tackle both temporal keyframe localization and question answering on videos. SeViLA framework consists of two modules: Localizer and Answerer, where both are parameter-efficiently fine-tuned from BLIP-2. We propose two ways of chaining these modules for cascaded inference and self-refinement. First, in the forward chain, the Localizer finds multiple language-aware keyframes in a video, which the Answerer uses to predict the answer. Second, in the reverse chain, the Answerer generates keyframe pseudo-labels to refine the Localizer, alleviating the need for expensive video moment localization annotations. Our SeViLA framework outperforms several strong baselines/previous works on five challenging video question answering and event prediction benchmarks, and achieves the state-of-the-art in both fine-tuning (NExT-QA and STAR) and zero-shot (NExT-QA, STAR, How2QA, and VLEP) settings. We show a comprehensive analysis of our framework, including the impact of Localizer, comparisons of Localizer with other temporal localization models, pre-training/self-refinement of Localizer, and varying the number of keyframes.

We present PERCEIVER-VL, a vision-and-language framework that efficiently handles high-dimensional multi- modal inputs such as long videos and text. Powered by the iterative latent-cross-attention of Perceiver, our framework scales with linear complexity, in contrast to the quadratic complexity of self-attention used in many state-of-the-art transformer-based models. To further improve the efficiency of our framework, we also study applying LayerDrop on cross-attention layers and introduce a mixed- stream architecture for cross-modal retrieval. We evaluate PERCEIVER-VL on diverse video-text and image-text benchmarks, where PERCEIVER-VL achieves the lowest GFLOPs and latency, while maintaining competitive performance. In addition, we also provide comprehensive analyses over various aspects of our framework, including pretraining data, scalability of latent size and input size, dropping cross-attention layers at inference to reduce latency, modality aggregation strategy, positional encoding, and weight initialization strategy.1

Recently, there has been an increasing interest in building question answering (QA) models that reason across multiple modalities, such as text and images. However, QA using images is often limited to just picking the answer from a pre-defined set of options. In addition, images in the real world, especially in news, have objects that are co-referential to the text, with complementary information from both modalities. In this paper, we present a new QA evaluation benchmark with 1,384 questions over news articles that require cross-media grounding of objects in images onto text. Specifically, the task involves multi-hop questions that require reasoning over image-caption pairs to identify the grounded visual object being referred to and then predicting a span from the news body text to answer the question. In addition, we introduce a novel multimedia data augmentation framework, based on cross-media knowledge extraction and synthetic question-answer generation, to automatically augment data that can provide weak supervision for this task. We evaluate both pipeline-based and end-to-end pretraining-based multimedia QA models on our benchmark, and show that they achieve promising performance, while considerably lagging behind human performance hence leaving large room for future work on this challenging new task.

Recently, fine-tuning language models pre-trained on large text corpora have provided huge improvements on vision-and-language (V&L) tasks as well as on pure language tasks. However, fine-tuning the entire parameter set of pre-trained models becomes impractical since the model size is growing rapidly. Hence, in this paper, we introduce adapter-based parameter-efficient transfer learning techniques to V&L models such as VL-BART and VL-T5. We evaluate our methods in a unified multi-task setup on both image-text and video-text benchmarks. For the image-text tasks, we use four diverse V&L datasets: VQAv2, GQA, NLVR2, and MSCOCO image captioning. For video-text tasks, we use TVQA, How2QA, TVC, and YC2C. With careful training and thorough experiments, we benchmark three popular adapter-based methods (Adapter, Hyperformer, Compacter) against the standard full fine-tuning and the recently proposed prompt-tuning approach. We also enhance the efficiency and performance of adapters by sharing their weights to attain knowledge across tasks. Our results demonstrate that training the adapter with the weight-sharing technique (4.18% of total parameters for image-text tasks and 3.39% for video-text tasks) can match the performance of fine-tuning the entire model. Lastly, we present a comprehensive analysis including the combination of adapter and task-specific prompts and the impact of V&L pre-training on adapters.11The code for our CVPR 2022 paper is available at: https://github.com/ylsung/VL_adapter.

Fine-tuning large pre-trained models on downstream tasks has been adopted in a variety of domains recently. However, it is costly to update the entire parameter set of large pre-trained models. Although recently proposed parameter-efficient transfer learning (PETL) techniques allow updating a small subset of parameters (e.g. only using 2% of parameters) inside a pre-trained backbone network for a new task, they only reduce the training memory requirement by up to 30%. This is because the gradient computation for the trainable parameters still requires back-propagation through the large pre-trained backbone model. To address this, we propose Ladder Side-Tuning (LST), a new PETL technique that can reduce training memory requirements by more substantial amounts. Unlike existing parameter-efficient methods that insert additional parameters inside backbone networks, we train a ladder side network, a small and separate network that takes intermediate activations as input via shortcut connections (ladders) from backbone networks and makes predictions. LST has significantly lower memory requirements than previous methods, because it does not require back-propagation through the backbone network, but instead only through the side network and ladder connections. We evaluate our method with various models (T5 and CLIP-T5) on both natural language processing (GLUE) and vision-and-language (VQA, GQA, NLVR2, MSCOCO) tasks. LST saves 69% of the memory costs to fine-tune the whole network, while other methods only save 26% of that in similar parameter usages (hence, 2.7x more memory savings). Moreover, LST achieves higher accuracy than Adapter and LoRA in a low-memory regime. To further show the advantage of this better memory efficiency, we also apply LST to larger T5 models (T5-large, T5-3B), attaining better GLUE performance than full fine-tuning and other PETL methods. The trend also holds in the experiments on vision-and-language tasks, where LST achieves similar accuracy to other PETL methods when training a similar number of parameters while also having 2.7x more memory savings. Our code is available at: https://github.com/ylsung/Ladder-Side-Tuning.

In this work, we present the Textless Vision-Language Transformer (TVLT), where homogeneous transformer blocks take raw visual and audio inputs for vision-and-language representation learning with minimal modality-specific design, and do not use text-specific modules such as tokenization or automatic speech recognition (ASR). TVLT is trained by reconstructing masked patches of continuous video frames and audio spectrograms (masked autoencoding) and contrastive modeling to align video and audio. TVLT attains performance comparable to its text-based counterpart on various multimodal tasks, such as visual question answering, image retrieval, video retrieval, and multimodal sentiment analysis, with 28x faster inference speed and only 1/3 of the parameters. Our findings suggest the possibility of learning compact and efficient visual-linguistic representations from low-level visual and audio signals without assuming the prior existence of text. Our code and checkpoints are available at: https://github.com/zinengtang/TVLT

Existing methods for vision-and-language learning typically require designing task-specific architectures and objectives for each task. For example, a multi-label answer classifier for visual question answering, a region scorer for referring expression comprehension, and a language decoder for image captioning, etc. To alleviate these hassles, in this work, we propose a unified framework that learns different tasks in a single architecture with the same language modeling objective, i.e., multimodal conditional text generation, where our models learn to generate labels in text based on the visual and textual inputs. On 7 popular vision-and-language benchmarks, including visual question answering, referring expression comprehension, visual commonsense reasoning, most of which have been previously modeled as discriminative tasks, our generative approach (with a single unified architecture) reaches comparable performance to recent task-specific state-of-the-art vision-and-language models. Moreover, our generative approach shows better generalization ability on questions that have rare answers. Also, we show that our framework allows multi-task learning in a single architecture with a single set of parameters, achieving similar performance to separately optimized single-task models. Our code is publicly available at: https://github.com/j-min/VL-T5

Since visual perception can give rich information beyond text descriptions for world understanding, there has been increasing interest in leveraging visual grounding for language learning. Recently, vokenization (Tan and Bansal, 2020) has attracted attention by using the predictions of a text-to-image retrieval model as labels for language model supervision. Despite its success, the method suffers from approximation error of using finite image labels and the lack of vocabulary diversity of a small image-text dataset. To overcome these limitations, we present VidLanKD, a video-language knowledge distillation method for improving language understanding. We train a multi-modal teacher model on a video-text dataset, and then transfer its knowledge to a student language model with a text dataset. To avoid approximation error, we propose to use different knowledge distillation objectives. In addition, the use of a large-scale video-text dataset helps learn diverse and richer vocabularies. In our experiments, VidLanKD achieves consistent improvements over text-only language models and vokenization models, on several downstream language understanding tasks including GLUE, SQuAD, and SWAG. We also demonstrate the improved world knowledge, physical reasoning, and temporal reasoning capabilities of our model by evaluating on the GLUE-diagnostics, PIQA, and TRACIE datasets. Lastly, we present comprehensive ablation studies as well as visualizations of the learned text-to-video grounding results of our teacher and student language models.

Mirroring the success of masked language models, vision-and-language counterparts like VILBERT, LXMERT and UNITER have achieved state of the art performance on a variety of multimodal discriminative tasks like visual question answering and visual grounding.Recent work has also successfully adapted such models towards the generative task of image captioning.This begs the question: Can these models go the other way and generate images from pieces of text?Our analysis of a popular representative from this model family -LXMERT -finds that it is unable to generate rich and semantically meaningful imagery with its current training setup.We introduce X-LXMERT, an extension to LXMERT with training refinements including: discretizing visual representations, using uniform masking with a large range of masking ratios and aligning the right pre-training datasets to the right objectives which enables it to paint.X-LXMERT's image generation capabilities rival state of the art generative models while its question answering and captioning abilities remains comparable to LXMERT.Finally, we demonstrate the generality of these training refinements by adding image generation capabilities into UNITER to produce X-UNITER.

Jaemin Cho, Minjoon Seo, Hannaneh Hajishirzi. Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP). 2019.

Yookoon Park, Jaemin Cho, Gunhee Kim. Proceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long Papers). 2018.