arXiv subjects

Svetlana Lazebnik

Publications and source records attributed to Svetlana Lazebnik.

At least 19 recordsLinked to original sources

arXiv2026

Constructive Distortion: Improving MLLMs with Attention-Guided Image Warping

Multimodal large language models (MLLMs) often miss small details and spatial relations in cluttered scenes, leading to errors in fine-grained perceptual grounding. We introduce AttWarp, a lightweight method that allocates more resolution to query-relevant content while compressing less informative areas, all while preserving global context. At test time, the approach uses an MLLM's cross-modal attention to perform rectilinear warping of the input image, reallocating spatial resolution toward regions the model deems important, without changing model weights or architecture. This attention-guided warping preserves all original image information but redistributes it non-uniformly, so small objects and subtle relationships become easier for the same model to read while the global layout remains intact. Across five benchmarks (TextVQA, GQA, DocVQA, POPE, MMMU) and four MLLMs (LLaVA, Qwen-VL, InternVL, and InstructBLIP), AttWarp consistently improves accuracy, strengthens compositional reasoning, and reduces hallucinations, outperforming four competitive baselines that manipulate raw images at test time. Together, these results show that attention-guided warping prioritizes information relevant to the query while preserving context, and that the same MLLMs perform better when given such warped inputs.

cs.CV↗

arXiv2026

Generalizable VLA Finetuning via Representation Anchoring and Language-Action Alignment

Finetuning a pretrained vision-language model (VLM) on robot demonstrations via behavior cloning (BC) has become the standard recipe for vision-language-action (VLA) policies. However, BC finetuning progressively overwrites the pretrained representations that support visual and semantic generalization. Co-training on web image-text data, a common remedy, applies language and action losses to separate observations, leaving VLAs with language-action misalignment that standard manipulation benchmarks do not expose. We propose Anchor-Align, which augments BC with two objectives: Vision-Language Anchoring distills layer-wise representations from a frozen VLM copy to prevent this drift, while Language-Action Alignment converts each action target into a discrete motion-direction label and jointly trains language and action prediction on the same robot observation. We conduct real-world evaluations across eight manipulation settings on single-arm xArm7 and bimanual YAM robots, using two VLA architectures with regression and flow-matching action heads. Across these settings, Anchor-Align consistently improves over BC on novel targets, layouts, and motion-sensitive bimanual tasks requiring coordinated control. At scale in simulation, we demonstrate consistent improvements on OOD perturbations, perceptual robustness, and long-horizon control across LIBERO-PRO, LIBERO-Plus, and CALVIN, respectively, suggesting that preserving pretrained representations and effective action learning are not fundamentally at odds. Project page: anchoralignvla.github.io

cs.RO↗

arXiv2026

ReaDiT Guidance: Control for Image and Video Generation using Diffusion Transformer Features

We present DiT Readout (ReaDiT) Guidance, a lightweight framework for controlling generation with Diffusion Transformer (DiT) models via their internal feature representations. ReaDiT Guidance uses features from a single DiT block to steer the generative process according to spatial targets - like depth, pose, or edge maps - provided at test time. Furthermore, since modern text-to-video models are largely built on DiT backbones, ReaDiT Guidance naturally extends to video generation, enabling camera and motion control. Experimental results demonstrate that our approach achieves competitive or improved results compared to existing feature-based and off-the-shelf adapter-based approaches while requiring fewer parameters.

cs.CV↗

arXiv2026

Learning Physics-Guided Residual Dynamics for Deformable Object Simulation

Simulating deformable objects is essential for a wide range of robotic manipulation applications, yet accurately predicting their dynamics remains challenging. We propose Physics-Guided Residual Dynamics (PGRD), a hybrid simulation framework that combines the advantages of physics-based and learning-based approaches. Specifically, PGRD combines an optimizable spring-mass simulator as a backbone with a learned neural network that predicts residual corrections to the physics-based predictions. We adopt a velocity-based formulation to ensure stable simulation and a sliding-window transformer architecture to capture temporal dependencies. We show that PGRD produces more accurate results than both purely physics-based and learning-based methods on a set of diverse real-world deformable objects. We further demonstrate the utility of PGRD in two applications: manipulation planning via Model Predictive Control, including a language-conditioned setting with a generated goal image; and interactive simulation via action-conditioned video prediction by 3D Gaussian Splatting.

cs.RO↗

arXiv2026

CRAFT: A Tendon-Driven Hand with Hybrid Hard-Soft Compliance

We introduce CRAFT hand, a tendon-driven anthropomorphic hand with hybrid hard-soft compliance for contact-rich manipulation. The design is based on a simple idea: contact is not uniform across the hand. Impacts concentrate at joints, while links carry most of the load. CRAFT places soft material at joints and keeps links rigid, and uses rollingcontact joint surfaces to keep flexion on repeatable motion paths. Fifteen motors mounted on the fingers drive the hand through tendons, keeping the form factor compact and the fingers light. In structural tests, CRAFT improves strength and endurance while maintaining comparable repeatability. In teleoperation, CRAFT improves handling of fragile and low-friction items, and the hand covers 33/33 grasps in the Feix taxonomy. The full design costs under $600 and will be released open-source with visionbased teleoperation and simulation integration. Project page: http://craft-hand.github.io/

cs.RO↗

arXiv2026

Robotic Manipulation by Imitating Generated Videos Without Physical Demonstrations

This work introduces Robots Imitating Generated Videos (RIGVid), a system that enables robots to perform complex manipulation tasks--such as pouring, wiping, and mixing--purely by imitating AI-generated videos, without requiring any physical demonstrations or robot-specific training. Given a language command and an initial scene image, a video diffusion model generates potential demonstration videos, and a vision-language model (VLM) automatically filters out results that do not follow the command. A 6D pose tracker then extracts object trajectories from the video, and the trajectories are retargeted to the robot in an embodiment-agnostic fashion. Through extensive real-world evaluations, we show that filtered generated videos are as effective as real demonstrations, and that performance improves with generation quality. We also show that relying on generated videos outperforms more compact alternatives such as keypoint prediction using VLMs, and that strong 6D pose tracking outperforms other ways to extract trajectories, such as dense feature point tracking. These findings suggest that videos produced by a state-of-the-art off-the-shelf model can offer an effective source of supervision for robotic manipulation.

cs.RO↗

arXiv2026

ZipLoRA: Any Subject in Any Style by Effectively Merging LoRAs

Methods for finetuning generative models for concept-driven personalization generally achieve strong results for subject-driven or style-driven generation. Recently, low-rank adaptations (LoRA) have been proposed as a parameter-efficient way of achieving concept-driven personalization. While recent work explores the combination of separate LoRAs to achieve joint generation of learned styles and subjects, existing techniques do not reliably address the problem; they often compromise either subject fidelity or style fidelity. We propose ZipLoRA, a method to cheaply and effectively merge independently trained style and subject LoRAs in order to achieve generation of any user-provided subject in any user-provided style. Experiments on a wide range of subject and style combinations show that ZipLoRA can generate compelling results with meaningful improvements over baselines in subject and style fidelity while preserving the ability to recontextualize. Project page: https://ziplora.github.io

cs.CV↗

arXiv2026

JoIN: Joint GANs Inversion for Intrinsic Image Decomposition

Intrinsic Image Decomposition (IID) is a challenging inverse problem that seeks to decompose a natural image into its underlying intrinsic components such as albedo and shading. While recent image decomposition methods rely on learning-based priors on these components, they often suffer from component cross-contamination owing to joint training of priors; or from Sim-to-Real gap since the priors trained on synthetic data are kept frozen during the inference on real images. In this work, we propose to solve the intrinsic image decomposition problem using a bank of Generative Adversarial Networks (GANs) as priors where each GAN is independently trained only on a single intrinsic component, providing stronger and more disentangled priors. At the core of our approach is the idea that the latent space of a GAN is a well-suited optimization domain to solve inverse problems. Given an input image, we propose to jointly invert the latent codes of a set of GANs and combine their outputs to reproduce the input. Contrary to all existing GAN inversion methods that are limited to inverting only a single GAN, our proposed approach, JoIN, is able to jointly invert multiple GANs using only a single image as supervision while still maintaining distribution priors of each intrinsic component. We show that our approach is modular, allowing various forward imaging models, and that it can successfully decompose both synthetic and real images. Further, taking inspiration from existing GAN inversion approaches, we allow for careful fine-tuning of the generator priors during the inference on real images. This way, our method is able to achieve excellent generalization on real images even though it uses only synthetic data to train the GAN priors. We demonstrate the success of our approach through exhaustive qualitative and quantitative evaluations and ablation studies on various datasets.

cs.CV↗

arXiv2025

UnZipLoRA: Separating Content and Style from a Single Image

This paper introduces UnZipLoRA, a method for decomposing an image into its constituent subject and style, represented as two distinct LoRAs (Low-Rank Adaptations). Unlike existing personalization techniques that focus on either subject or style in isolation, or require separate training sets for each, UnZipLoRA disentangles these elements from a single image by training both the LoRAs simultaneously. UnZipLoRA ensures that the resulting LoRAs are compatible, i.e., they can be seamlessly combined using direct addition. UnZipLoRA enables independent manipulation and recontextualization of subject and style, including generating variations of each, applying the extracted style to new subjects, and recombining them to reconstruct the original image or create novel variations. To address the challenge of subject and style entanglement, UnZipLoRA employs a novel prompt separation technique, as well as column and block separation strategies to accurately preserve the characteristics of subject and style, and ensure compatibility between the learned LoRAs. Evaluation with human studies and quantitative metrics demonstrates UnZipLoRA's effectiveness compared to other state-of-the-art methods, including DreamBooth-LoRA, Inspiration Tree, and B-LoRA.

cs.CV↗

arXiv2025

A Real-to-Sim-to-Real Approach to Robotic Manipulation with VLM-Generated Iterative Keypoint Rewards

Task specification for robotic manipulation in open-world environments is challenging, requiring flexible and adaptive objectives that align with human intentions and can evolve through iterative feedback. We introduce Iterative Keypoint Reward (IKER), a visually grounded, Python-based reward function that serves as a dynamic task specification. Our framework leverages VLMs to generate and refine these reward functions for multi-step manipulation tasks. Given RGB-D observations and free-form language instructions, we sample keypoints in the scene and generate a reward function conditioned on these keypoints. IKER operates on the spatial relationships between keypoints, leveraging commonsense priors about the desired behaviors, and enabling precise SE(3) control. We reconstruct real-world scenes in simulation and use the generated rewards to train reinforcement learning (RL) policies, which are then deployed into the real world-forming a real-to-sim-to-real loop. Our approach demonstrates notable capabilities across diverse scenarios, including both prehensile and non-prehensile tasks, showcasing multi-step task execution, spontaneous error recovery, and on-the-fly strategy adjustments. The results highlight IKER's effectiveness in enabling robots to perform multi-step tasks in dynamic environments through iterative reward shaping.

cs.RO↗

arXiv2024

Street TryOn: Learning In-the-Wild Virtual Try-On from Unpaired Person Images

Most virtual try-on research is motivated to serve the fashion business by generating images to demonstrate garments on studio models at a lower cost. However, virtual try-on should be a broader application that also allows customers to visualize garments on themselves using their own casual photos, known as in-the-wild try-on. Unfortunately, the existing methods, which achieve plausible results for studio try-on settings, perform poorly in the in-the-wild context. This is because these methods often require paired images (garment images paired with images of people wearing the same garment) for training. While such paired data is easy to collect from shopping websites for studio settings, it is difficult to obtain for in-the-wild scenes. In this work, we fill the gap by (1) introducing a StreetTryOn benchmark to support in-the-wild virtual try-on applications and (2) proposing a novel method to learn virtual try-on from a set of in-the-wild person images directly without requiring paired data. We tackle the unique challenges, including warping garments to more diverse human poses and rendering more complex backgrounds faithfully, by a novel DensePose warping correction method combined with diffusion-based conditional inpainting. Our experiments show competitive performance for standard studio try-on tasks and SOTA performance for street try-on and cross-domain try-on tasks.

cs.CV↗

arXiv2024

Shadows Don't Lie and Lines Can't Bend! Generative Models don't know Projective Geometry...for now

Generative models can produce impressively realistic images. This paper demonstrates that generated images have geometric features different from those of real images. We build a set of collections of generated images, prequalified to fool simple, signal-based classifiers into believing they are real. We then show that prequalified generated images can be identified reliably by classifiers that only look at geometric properties. We use three such classifiers. All three classifiers are denied access to image pixels, and look only at derived geometric features. The first classifier looks at the perspective field of the image, the second looks at lines detected in the image, and the third looks at relations between detected objects and shadows. Our procedure detects generated images more reliably than SOTA local signal based detectors, for images from a number of distinct generators. Saliency maps suggest that the classifiers can identify geometric problems reliably. We conclude that current generators cannot reliably reproduce geometric properties of real images.

cs.CV↗

arXiv2023

MultiStyleGAN: Multiple One-shot Image Stylizations using a Single GAN

Image stylization aims at applying a reference style to arbitrary input images. A common scenario is one-shot stylization, where only one example is available for each reference style. Recent approaches for one-shot stylization such as JoJoGAN fine-tune a pre-trained StyleGAN2 generator on a single style reference image. However, such methods cannot generate multiple stylizations without fine-tuning a new model for each style separately. In this work, we present a MultiStyleGAN method that is capable of producing multiple different stylizations at once by fine-tuning a single generator. The key component of our method is a learnable transformation module called Style Transformation Network. It takes latent codes as input, and learns linear mappings to different regions of the latent space to produce distinct codes for each style, resulting in a multistyle space. Our model inherently mitigates overfitting since it is trained on multiple styles, hence improving the quality of stylizations. Our method can learn upwards of $120$ image stylizations at once, bringing $8\times$ to $60\times$ improvement in training time over recent competing methods. We support our results through user studies and quantitative results that indicate meaningful improvements over existing methods.

cs.CV↗

arXiv2023

One-Shot Stylization for Full-Body Human Images

The goal of human stylization is to transfer full-body human photos to a style specified by a single art character reference image. Although previous work has succeeded in example-based stylization of faces and generic scenes, full-body human stylization is a more complex domain. This work addresses several unique challenges of stylizing full-body human images. We propose a method for one-shot fine-tuning of a pose-guided human generator to preserve the "content" (garments, face, hair, pose) of the input photo and the "style" of the artistic reference. Since body shape deformation is an essential component of an art character's style, we incorporate a novel skeleton deformation module to reshape the pose of the input person and modify the DiOr pose-guided person generator to be more robust to the rescaled poses falling outside the distribution of the realistic poses that the generator is originally trained on. Several human studies verify the effectiveness of our approach.

cs.CV↗

arXiv2022

Robust Online Video Instance Segmentation with Track Queries

Recently, transformer-based methods have achieved impressive results on Video Instance Segmentation (VIS). However, most of these top-performing methods run in an offline manner by processing the entire video clip at once to predict instance mask volumes. This makes them incapable of handling the long videos that appear in challenging new video instance segmentation datasets like UVO and OVIS. We propose a fully online transformer-based video instance segmentation model that performs comparably to top offline methods on the YouTube-VIS 2019 benchmark and considerably outperforms them on UVO and OVIS. This method, called Robust Online Video Segmentation (ROVIS), augments the Mask2Former image instance segmentation model with track queries, a lightweight mechanism for carrying track information from frame to frame, originally introduced by the TrackFormer method for multi-object tracking. We show that, when combined with a strong enough image segmentation architecture, track queries can exhibit impressive accuracy while not being constrained to short videos.

cs.CV↗

arXiv2022

Dressing in Order: Recurrent Person Image Generation for Pose Transfer, Virtual Try-on and Outfit Editing

We propose a flexible person generation framework called Dressing in Order (DiOr), which supports 2D pose transfer, virtual try-on, and several fashion editing tasks. The key to DiOr is a novel recurrent generation pipeline to sequentially put garments on a person, so that trying on the same garments in different orders will result in different looks. Our system can produce dressing effects not achievable by existing work, including different interactions of garments (e.g., wearing a top tucked into the bottom or over it), as well as layering of multiple garments of the same type (e.g., jacket over shirt over t-shirt). DiOr explicitly encodes the shape and texture of each garment, enabling these elements to be edited separately. Joint training on pose transfer and inpainting helps with detail preservation and coherence of generated garments. Extensive evaluations show that DiOr outperforms other recent methods like ADGAN in terms of output quality, and handles a wide range of editing functions for which there is no direct supervision.

cs.CV↗

arXiv2022

Transfer of Representations to Video Label Propagation: Implementation Factors Matter

This work studies feature representations for dense label propagation in video, with a focus on recently proposed methods that learn video correspondence using self-supervised signals such as colorization or temporal cycle consistency. In the literature, these methods have been evaluated with an array of inconsistent settings, making it difficult to discern trends or compare performance fairly. Starting with a unified formulation of the label propagation algorithm that encompasses most existing variations, we systematically study the impact of important implementation factors in feature extraction and label propagation. Along the way, we report the accuracies of properly tuned supervised and unsupervised still image baselines, which are higher than those found in previous works. We also demonstrate that augmenting video-based correspondence cues with still-image-based ones can further improve performance. We then attempt a fair comparison of recent video-based methods on the DAVIS benchmark, showing convergence of best methods to performance levels near our strong ImageNet baseline, despite the usage of a variety of specialized video-based losses and training particulars. Additional comparisons on JHMDB and VIP datasets confirm the similar performance of current methods. We hope that this study will help to improve evaluation practices and better inform future research directions in temporal correspondence.

cs.CV↗

arXiv2021

Bridging the Imitation Gap by Adaptive Insubordination

In practice, imitation learning is preferred over pure reinforcement learning whenever it is possible to design a teaching agent to provide expert supervision. However, we show that when the teaching agent makes decisions with access to privileged information that is unavailable to the student, this information is marginalized during imitation learning, resulting in an "imitation gap" and, potentially, poor results. Prior work bridges this gap via a progression from imitation learning to reinforcement learning. While often successful, gradual progression fails for tasks that require frequent switches between exploration and memorization. To better address these tasks and alleviate the imitation gap we propose 'Adaptive Insubordination' (ADVISOR). ADVISOR dynamically weights imitation and reward-based reinforcement learning losses during training, enabling on-the-fly switching between imitation and exploration. On a suite of challenging tasks set within gridworlds, multi-agent particle environments, and high-fidelity 3D simulators, we show that on-the-fly switching with ADVISOR outperforms pure imitation, pure reinforcement learning, as well as their sequential and parallel combinations.

cs.LG↗