Saman Motamed

Chen Henry Wu, Saman Motamed, Shaunak Srivastava et al. · cmu

Generative models (e.g., GANs, diffusion models) learn the underlying data distribution in an unsupervised manner. However, many applications of interest require sampling from a particular region of the output space or sampling evenly over a range of characteristics. For efficient sampling in these scenarios, we propose Generative Visual Prompt (PromptGen), a framework for distributional control over pre-trained generative models by incorporating knowledge of other off-the-shelf models. PromptGen defines control as energy-based models (EBMs) and samples images in a feed-forward manner by approximating the EBM with invertible neural networks, avoiding optimization at inference. Our experiments demonstrate how PromptGen can efficiently sample from several unconditional generative models (e.g., StyleGAN2, StyleNeRF, diffusion autoencoder, NVAE) in a controlled or/and de-biased manner using various off-the-shelf models: (1) with the CLIP model as control, PromptGen can sample images guided by text, (2) with image classifiers as control, PromptGen can de-bias generative models across a set of attributes or attribute combinations, and (3) with inverse graphics models as control, PromptGen can sample images of the same identity in different poses. (4) Finally, PromptGen reveals that the CLIP model shows a "reporting bias" when used as control, and PromptGen can further de-bias this controlled distribution in an iterative manner. The code is available at https://github.com/ChenWu98/Generative-Visual-Prompt.

Saman Motamed, Jianjin Xu, Chen Henry Wu et al.

Generative models such as StyleGAN2 and Stable Diffusion have achieved state-of-the-art performance in computer vision tasks such as image synthesis, inpainting, and de-noising. However, current generative models for face inpainting often fail to preserve fine facial details and the identity of the person, despite creating aesthetically convincing image structures and textures. In this work, we propose Person Aware Tuning (PAT) of Mask-Aware Transformer (MAT) for face inpainting, which addresses this issue. Our proposed method, PATMAT, effectively preserves identity by incorporating reference images of a subject and fine-tuning a MAT architecture trained on faces. By using ~40 reference images, PATMAT creates anchor points in MAT's style module, and tunes the model using the fixed anchors to adapt the model to a new face identity. Moreover, PATMAT's use of multiple images per anchor during training allows the model to use fewer reference images than competing methods. We demonstrate that PATMAT outperforms state-of-the-art models in terms of image quality, the preservation of person-specific details, and the identity of the subject. Our results suggest that PATMAT can be a promising approach for improving the quality of personalized face inpainting.

Saman Motamed, Danda Pani Paudel, Luc Van Gool

Text-to-Image (T2I) models excel at synthesizing concepts such as nouns, appearances, and styles. To enable customized content creation based on a few example images of a concept, methods such as Textual Inversion and DreamBooth invert the desired concept and enable synthesizing it in new scenes. However, inverting personalized concepts that go beyond object appearance and style (adjectives and verbs) through natural language remains a challenge. Two key characteristics of these concepts contribute to the limitations of current inversion methods. 1) Adjectives and verbs are entangled with nouns (subject) and can hinder appearance-based inversion methods, where the subject appearance leaks into the concept embedding, and 2) describing such concepts often extends beyond single word embeddings. In this study, we introduce Lego, a textual inversion method designed to invert subject-entangled concepts from a few example images. Lego disentangles concepts from their associated subjects using a simple yet effective Subject Separation step and employs a Context Loss that guides the inversion of single/multi-embedding concepts. In a thorough user study, Lego-generated concepts were preferred over 70% of the time when compared to the baseline in terms of authentically generating concepts according to a reference. Additionally, visual question answering using an LLM suggested Lego-generated concepts are better aligned with the text description of the concept.

Saman Motamed, Laura Culp, Kevin Swersky et al.

AI video generation is undergoing a revolution, with quality and realism advancing rapidly. These advances have led to a passionate scientific debate: Do video models learn "world models" that discover laws of physics -- or, alternatively, are they merely sophisticated pixel predictors that achieve visual realism without understanding the physical principles of reality? We address this question by developing Physics-IQ, a comprehensive benchmark dataset that can only be solved by acquiring a deep understanding of various physical principles, like fluid dynamics, optics, solid mechanics, magnetism and thermodynamics. We find that across a range of current models (Sora, Runway, Pika, Lumiere, Stable Video Diffusion, and VideoPoet), physical understanding is severely limited, and unrelated to visual realism. At the same time, some test cases can already be successfully solved. This indicates that acquiring certain physical principles from observation alone may be possible, but significant challenges remain. While we expect rapid advances ahead, our work demonstrates that visual realism does not imply physical understanding. Our project page is at https://physics-iq.github.io; code at https://github.com/google-deepmind/physics-IQ-benchmark.

Zuhao Liu, Aleksandar Yanev, Ahmad Mahmood et al.

Advances in video generation have significantly improved the realism and quality of created scenes. This has fueled interest in developing intuitive tools that let users leverage video generation as world simulators. Text-to-video (T2V) generation is one such approach, enabling video creation from text descriptions only. Yet, due to the inherent ambiguity in texts and the limited temporal information offered by text prompts, researchers have explored additional control signals like trajectory-guided systems, for more accurate T2V generation. Nonetheless, methods to evaluate whether T2V models can generate realistic interactions between multiple objects are lacking. We introduce InTraGen, a pipeline for improved trajectory-based generation of object interaction scenarios. We propose 4 new datasets and a novel trajectory quality metric to evaluate the performance of the proposed InTraGen. To achieve object interaction, we introduce a multi-modal interaction encoding pipeline with an object ID injection mechanism that enriches object-environment interactions. Our results demonstrate improvements in both visual fidelity and quantitative performance. Code and datasets are available at https://github.com/insait-institute/InTraGen

Jianjin Xu, Saman Motamed, Praneetha Vaddamanu et al.

Face inpainting is important in various applications, such as photo restoration, image editing, and virtual reality. Despite the significant advances in face generative models, ensuring that a person's unique facial identity is maintained during the inpainting process is still an elusive goal. Current state-of-the-art techniques, exemplified by MyStyle, necessitate resource-intensive fine-tuning and a substantial number of images for each new identity. Furthermore, existing methods often fall short in accommodating user-specified semantic attributes, such as beard or expression. To improve inpainting results, and reduce the computational complexity during inference, this paper proposes the use of Parallel Visual Attention (PVA) in conjunction with diffusion models. Specifically, we insert parallel attention matrices to each cross-attention module in the denoising network, which attends to features extracted from reference images by an identity encoder. We train the added attention modules and identity encoder on CelebAHQ-IDI, a dataset proposed for identity-preserving face inpainting. Experiments demonstrate that PVA attains unparalleled identity resemblance in both face inpainting and face inpainting with language guidance tasks, in comparison to various benchmarks, including MyStyle, Paint by Example, and Custom Diffusion. Our findings reveal that PVA ensures good identity preservation while offering effective language-controllability. Additionally, in contrast to Custom Diffusion, PVA requires just 40 fine-tuning steps for each new identity, which translates to a significant speed increase of over 20 times.

Reni Paskaleva, Mykyta Holubakha, Andela Ilic et al.

Canonical emotions, such as happy, sad, and fearful, are easy to understand and annotate. However, emotions are often compound, e.g. happily surprised, and can be mapped to the action units (AUs) used for expressing emotions, and trivially to the canonical ones. Intuitively, emotions are continuous as represented by the arousal-valence (AV) model. An interpretable unification of these four modalities - namely, Canonical, Compound, AUs, and AV - is highly desirable, for a better representation and understanding of emotions. However, such unification remains to be unknown in the current literature. In this work, we propose an interpretable and unified emotion model, referred as C2A2. We also develop a method that leverages labels of the non-unified models to annotate the novel unified one. Finally, we modify the text-conditional diffusion models to understand continuous numbers, which are then used to generate continuous expressions using our unified emotion model. Through quantitative and qualitative experiments, we show that our generated images are rich and capture subtle expressions. Our work allows a fine-grained generation of expressions in conjunction with other textual inputs and offers a new label space for emotions at the same time.

Lin Zhang, Linghan Xu, Saman Motamed et al.

Unsupervised domain adaptation (UDA) for image classification has made remarkable progress in transferring classification knowledge from a labeled source domain to an unlabeled target domain, thanks to effective domain alignment techniques. Recently, in order to further improve performance on a target domain, many Single-Target Active Domain Adaptation (ST-ADA) methods have been proposed to identify and annotate the salient and exemplar target samples. However, it requires one model to be trained and deployed for each target domain and the domain label associated with each test sample. This largely restricts its application in the ubiquitous scenarios with multiple target domains. Therefore, we propose a Multi-Target Active Domain Adaptation (MT-ADA) framework for image classification, named D3GU, to simultaneously align different domains and actively select samples from them for annotation. This is the first research effort in this field to our best knowledge. D3GU applies Decomposed Domain Discrimination (D3) during training to achieve both source-target and target-target domain alignments. Then during active sampling, a Gradient Utility (GU) score is designed to weight every unlabeled target image by its contribution towards classification and domain alignment tasks, and is further combined with KMeans clustering to form GU-KMeans for diverse image sampling. Extensive experiments on three benchmark datasets, Office31, OfficeHome, and DomainNet, have been conducted to validate consistently superior performance of D3GU for MT-ADA.

Saman Motamed, William Harvey, Benjamin Klein et al.

Existing video object removal methods excel at inpainting content "behind" the object and correcting appearance-level artifacts such as shadows and reflections. However, when the removed object has more significant interactions, such as collisions with other objects, current models fail to correct them and produce implausible results. We present VOID, a video object removal framework designed to perform physically-plausible inpainting in these complex scenarios. To train the model, we generate a new paired dataset of counterfactual object removals using Kubric and HUMOTO, where removing an object requires altering downstream physical interactions. During inference, a vision-language model identifies regions of the scene affected by the removed object. These regions are then used to guide a video diffusion model that generates physically consistent counterfactual outcomes. Experiments on both synthetic and real data show that our approach better preserves consistent scene dynamics after object removal compared to prior video object removal methods. We hope this framework sheds light on how to make video editing models better simulators of the world through high-level causal reasoning.

Saman Motamed, Wouter Van Gansbeke, Luc Van Gool

With recent advances in image and video diffusion models for content creation, a plethora of techniques have been proposed for customizing their generated content. In particular, manipulating the cross-attention layers of Text-to-Image (T2I) diffusion models has shown great promise in controlling the shape and location of objects in the scene. Transferring image-editing techniques to the video domain, however, is extremely challenging as object motion and temporal consistency are difficult to capture accurately. In this work, we take a first look at the role of cross-attention in Text-to-Video (T2V) diffusion models for zero-shot video editing. While one-shot models have shown potential in controlling motion and camera movement, we demonstrate zero-shot control over object shape, position and movement in T2V models. We show that despite the limitations of current T2V models, cross-attention guidance can be a promising approach for editing videos.

Saman Motamed, Minghao Chen, Luc Van Gool et al.

Despite impressive visual fidelity, modern video generative models frequently produce sequences that violate intuitive physical laws, such as objects floating, teleporting, or morphing in ways that defy causality. While humans can easily detect such implausibilities, there remains no robust method for quantitatively assessing physical realism in video. In this work, we explore whether Video-Language Models (VLMs) can be trained to serve as reliable judges of physical plausibility. We find that existing VLMs struggle to identify physics violations, exposing fundamental limitations in their temporal and causal reasoning. To address this, we introduce TRAVL, a fine-tuning recipe that combines a balanced training dataset with a trajectory-aware attention module to improve motion encoding and discrimination in VLMs. To evaluate physical reasoning more rigorously, we propose ImplausiBench, a benchmark of 300 videos (150 real, 150 generated) that removes linguistic biases and isolates visual-temporal understanding. Performance is reported both with gold-standard human judgments and stricter LLM-as-judge metrics. Together, TRAVL and ImplausiBench offer a unified framework for probing and improving physical plausibility in multimodal models, shedding light on a challenging and underexplored aspect of visual-temporal understanding.

Saman Motamed, Farzad Khalvati

Generative Adversarial Networks can learn the mapping of random noise to realistic images in a semi-supervised framework. This mapping ability can be used for semi-supervised image classification to detect images of an unknown class where there is no training data to be used for supervised classification. However, if the unknown class shares similar characteristics to the known class(es), GANs can learn to generalize and generate images that look like both classes. This generalization ability can hinder the classification performance. In this work, we propose the Vanishing Twin GAN. By training a weak GAN and using its generated output image parallel to the regular GAN, the Vanishing Twin training improves semi-supervised image classification where image similarity can hurt classification tasks.

Saman Motamed, Farzad Khalvati

From generating never-before-seen images to domain adaptation, applications of Generative Adversarial Networks (GANs) spread wide in the domain of vision and graphics problems. With the remarkable ability of GANs in learning the distribution and generating images of a particular class, they can be used for semi-supervised classification tasks. However, the problem is that if two classes of images share similar characteristics, the GAN might learn to generalize and hinder the classification of the two classes. In this paper, we use various images from MNIST and Fashion-MNIST datasets to illustrate how similar images cause the GAN to generalize, leading to the poor classification of images. We propose a modification to the traditional training of GANs that allows for improved multi-class classification in similar classes of images in a semi-supervised learning framework.

Saman Motamed, Patrik Rogalla, Farzad Khalvati

COVID-19 spread across the globe at an immense rate has left healthcare systems incapacitated to diagnose and test patients at the needed rate. Studies have shown promising results for detection of COVID-19 from viral bacterial pneumonia in chest X-rays. Automation of COVID-19 testing using medical images can speed up the testing process of patients where health care systems lack sufficient numbers of the reverse-transcription polymerase chain reaction (RT-PCR) tests. Supervised deep learning models such as convolutional neural networks (CNN) need enough labeled data for all classes to correctly learn the task of detection. Gathering labeled data is a cumbersome task and requires time and resources which could further strain health care systems and radiologists at the early stages of a pandemic such as COVID-19. In this study, we propose a randomized generative adversarial network (RANDGAN) that detects images of an unknown class (COVID-19) from known and labelled classes (Normal and Viral Pneumonia) without the need for labels and training data from the unknown class of images (COVID-19). We used the largest publicly available COVID-19 chest X-ray dataset, COVIDx, which is comprised of Normal, Pneumonia, and COVID-19 images from multiple public databases. In this work, we use transfer learning to segment the lungs in the COVIDx dataset. Next, we show why segmentation of the region of interest (lungs) is vital to correctly learn the task of classification, specifically in datasets that contain images from different resources as it is the case for the COVIDx dataset. Finally, we show improved results in detection of COVID-19 cases using our generative model (RANDGAN) compared to conventional generative adversarial networks (GANs) for anomaly detection in medical images, improving the area under the ROC curve from 0.71 to 0.77.

Saman Motamed, Patrik Rogalla, Farzad Khalvati

Successful training of convolutional neural networks (CNNs) requires a substantial amount of data. With small datasets networks generalize poorly. Data Augmentation techniques improve the generalizability of neural networks by using existing training data more effectively. Standard data augmentation methods, however, produce limited plausible alternative data. Generative Adversarial Networks (GANs) have been utilized to generate new data and improve the performance of CNNs. Nevertheless, data augmentation techniques for training GANs are under-explored compared to CNNs. In this work, we propose a new GAN architecture for augmentation of chest X-rays for semi-supervised detection of pneumonia and COVID-19 using generative models. We show that the proposed GAN can be used to effectively augment data and improve classification accuracy of disease in chest X-rays for pneumonia and COVID-19. We compare our augmentation GAN model with Deep Convolutional GAN and traditional augmentation methods (rotate, zoom, etc) on two different X-ray datasets and show our GAN-based augmentation method surpasses other augmentation methods for training a GAN in detecting anomalies in X-ray images.

Saman Motamed, Isha Gujrathi, Dominik Deniffel et al.

The segmentation of prostate whole gland and transition zone in Diffusion Weighted MRI (DWI) are the first step in designing computer-aided detection algorithms for prostate cancer. However, variations in MRI acquisition parameters and scanner manufacturing result in different appearances of prostate tissue in the images. Convolutional neural networks (CNNs) which have shown to be successful in various medical image analysis tasks including segmentation are typically sensitive to the variations in imaging parameters. This sensitivity leads to poor segmentation performance of CNNs trained on a source cohort and tested on a target cohort from a different scanner and hence, it limits the applicability of CNNs for cross-cohort training and testing. Contouring prostate whole gland and transition zone in DWI images are time-consuming and expensive. Thus, it is important to enable CNNs pretrained on images of source domain, to segment images of target domain with minimum requirement for manual segmentation of images from the target domain. In this work, we propose a transfer learning method based on a modified U-net architecture and loss function, for segmentation of prostate whole gland and transition zone in DWIs using a CNN pretrained on a source dataset and tested on the target dataset. We explore the effect of the size of subset of target dataset used for fine-tuning the pre-trained CNN on the overall segmentation accuracy. Our results show that with a fine-tuning data as few as 30 patients from the target domain, the proposed transfer learning-based algorithm can reach dice score coefficient of 0.80 for both prostate whole gland and transition zone segmentation. Using a fine-tuning data of 115 patients from the target domain, dice score coefficient of 0.85 and 0.84 are achieved for segmentation of whole gland and transition zone, respectively, in the target domain.