We introduce a simple yet effective technique for estimating lighting from a single low-dynamic-range (LDR) image by reframing the task as a chrome ball inpainting problem. This approach leverages a pre-trained diffusion model, Stable Diffusion XL, to overcome the generalization failures of existing methods that rely on limited HDR panorama datasets. While conceptually simple, the task remains challenging because diffusion models often insert incorrect or inconsistent content and cannot readily generate chrome balls in HDR format. Our analysis reveals that the inpainting process is highly sensitive to the initial noise in the diffusion process, occasionally resulting in unrealistic outputs. To address this, we first introduce DiffusionLight (Phongthawee et al. 2024), which uses iterative inpainting to compute a median chrome ball from multiple outputs to serve as a stable, low-frequency lighting prior that guides the generation of a high-quality final result. To generate high-dynamic-range (HDR) light probes, an Exposure LoRA is fine-tuned to create LDR images at multiple exposure values, which are then merged. While effective, DiffusionLight is time-intensive, requiring approximately 30 minutes per estimation. To reduce this overhead, we introduce DiffusionLight-Turbo, which reduces the runtime to about 30 seconds with minimal quality loss. This 60x speedup is achieved by training a Turbo LoRA to directly predict the averaged chrome balls from the iterative process. Inference is further streamlined into a single denoising pass using a LoRA swapping technique. Experimental results that show our method produces convincing light estimates across diverse settings and demonstrates superior generalization to in-the-wild scenarios.

We introduce a novel approach to single-view face relighting in the wild, addressing challenges such as global illumination and cast shadows. A common scheme in recent methods involves intrinsicallydecomposing an input image into 3D shape, albedo, and lighting, then recomposing it with the target lighting. However, estimating these components is error-prone and requires many training examples with ground-truth lighting to generalize well. Our work bypasses the need for accurate intrinsic estimation and can be trained solely on 2D images without any light stage data, relit pairs, multi-view images, or lighting ground truth. Our key idea is to leverage a conditional diffusion implicit model (DDIM) for decoding a disentangled light encoding along with other encodings related to 3D shape and facial identity inferred from off-the-shelf estimators. We propose a novel conditioning technique that simplifies modeling the complex interaction between light and geometry. It uses a rendered shading reference along with a shadow map, inferred using a simple and effective technique, to spatially modulate the DDIM. Moreover, we propose a single-shot relighting framework that requires just one network pass, given pre-processed data, and even outperforms the teacher model across all metrics. Our method realistically relights in-the-wild images with temporally consistent cast shadows under varying lighting conditions. We achieve state-of-the-art performance on the standard benchmark Multi-PIE and rank highest in user studies. Please visit our page: https://diffusion-face-relighting-pp.github.io.

One practical approach to infer 3D scene structure from a single image is to retrieve a closely matching 3D model from a database and align it with the object in the image. Existing methods rely on supervised training with images and pose annotations, which limits them to a narrow set of object categories. To address this, we propose a weakly supervised 9-DoF alignment method for inexact 3D models that requires no pose annotations and generalizes to unseen categories. Our approach derives a novel feature space based on foundation features that ensure multi-view consistency and overcome symmetry ambiguities inherent in foundation features using a self-supervised triplet loss. Additionally, we introduce a texture-invariant pose refinement technique that performs dense alignment in normalized object coordinates, estimated through the enhanced feature space. We conduct extensive evaluations on the real-world ScanNet25k dataset, where our method outperforms SOTA weakly supervised baselines by +4.3% mean alignment accuracy and is the only weakly supervised approach to surpass the supervised ROCA by +2.7%. To assess generalization, we introduce SUN2CAD, a real-world test set with 20 novel object categories, where our method achieves SOTA results without prior training on them.

While diffusion models show promising results in image editing given a target prompt, achieving both prompt fidelity and background preservation remains difficult. Recent works have introduced score distillation techniques that leverage the rich generative prior of text-to-image diffusion models to solve this task without additional fine-tuning. However, these methods often struggle with tasks such as object insertion. Our investigation of these failures reveals significant variations in gradient magnitude and spatial distribution, making hyperparameter tuning highly input-specific or unsuccessful. To address this, we propose two simple yet effective modifications: attention-based spatial regularization and gradient filtering-normalization, both aimed at reducing these variations during gradient updates. Experimental results show our method outperforms state-of-the-art score distillation techniques in prompt fidelity, improving successful edits while preserving the background. Users also preferred our method over state-of-the-art techniques across three metrics, and by 58-64% overall.

We propose Diffusion Noise Optimization (DNO), a new method that effectively leverages existing motion diffusion models as motion priors for a wide range of motion-related tasks. Instead of training a task-specific diffusion model for each new task, DNO operates by optimizing the diffusion latent noise of an existing pre-trained text-to-motion model. Given the corresponding latent noise of a human motion, it propagates the gradient from the target criteria defined on the motion space through the whole denoising process to update the diffusion latent noise. As a result, DNO supports any use cases where criteria can be defined as a function of motion. In particular, we show that, for motion editing and control, DNO outperforms existing meth-ods in both achieving the objective and preserving the motion content. DNO accommodates a diverse range of editing modes, including changing trajectory, pose, joint lo-cations, or avoiding newly added obstacles. In addition, DNO is effective in motion denoising and completion, pro-ducing smooth and realistic motion from noisy and partial inputs. DNO achieves these results at inference time with-out the need for model retraining, offering great versatility for any defined reward or loss function on the motion rep-resentation.

We present a simple yet effective technique to estimate lighting in a single input image. Current techniques rely heavily on HDR panorama datasets to train neural networks to regress an input with limited field-of-view to a full environment map. However, these approaches often struggle with real-world, uncontrolled settings due to the limited diversity and size of their datasets. To address this problem, we leverage diffusion models trained on billions of standard images to render a chrome ball into the input image. Despite its simplicity, this task remains challenging: the diffusion models often insert incorrect or inconsistent objects and cannot readily generate chrome balls in HDR format. Our research uncovers a surprising relationship between the appearance of chrome balls and the initial diffusion noise map, which we utilize to consistently generate high-quality chrome balls. We further fine-tune an LDR diffusion model (Stable Diffusion XL) with LoRA, enabling it to perform exposure bracketing for HDR light estimation. Our method produces convincing light estimates across diverse settings and demonstrates superior generalization to in-the-wild scenarios.

Despite the remarkable success of diffusion models in image generation, slow sampling remains a persistent issue. To accelerate the sampling process, prior studies have reformulated diffusion sampling as an ODE/SDE and introduced higher-order numerical methods. However, these methods often produce divergence artifacts, especially with a low number of sampling steps, which limits the achievable acceleration. In this paper, we investigate the potential causes of these artifacts and suggest that the small stability regions of these methods could be the principal cause. To address this issue, we propose two novel techniques. The first technique involves the incorporation of Heavy Ball (HB) momentum, a well-known technique for improving optimization, into existing diffusion numerical methods to expand their stability regions. We also prove that the resulting methods have first-order convergence. The second technique, called Generalized Heavy Ball (GHVB), constructs a new high-order method that offers a variable trade-off between accuracy and artifact suppression. Experimental results show that our techniques are highly effective in reducing artifacts and improving image quality, surpassing state-of-the-art diffusion solvers on both pixel-based and latent-based diffusion models for low-step sampling. Our research provides novel insights into the design of numerical methods for future diffusion work.

Cross-modal training using 2D-3D paired datasets, such as those containing multi-view images and 3D scene scans, presents an effective way to enhance 2D scene understanding by introducing geometric and view-invariance priors into 2D features. However, the need for large-scale scene datasets can impede scalability and further improvements. This paper explores an alternative learning method by leveraging a lightweight and publicly available type of 3D data in the form of CAD models. We construct a 3D space with geometric-aware alignment where the similarity in this space reflects the geometric similarity of CAD models based on the Chamfer distance. The acquired geometric-aware properties are then induced into 2D features, which boost performance on downstream tasks more effectively than existing RGB-CAD approaches. Our technique is not limited to paired RGB-CAD datasets. By training exclu-sively on pseudo pairs generated from CAD-based reconstruction methods, we enhance the performance of SOTA 2D pretrained models that use ResNet-50 or ViT-B back-bones on various 2D understanding tasks. We also achieve comparable results to SOTA methods trained on scene scans on four tasks in NYUv2, SUNRGB-D, indoor ADE20k, and indoor/outdoor COCO, despite using lightweight CAD models or pseudo data. Please visit our page: https://GeoAware2dRepUsingCAD.github.io/

Our paper seeks to transfer the hairstyle of a reference image to an input photo for virtual hair tryon. We target a variety of challenges scenarios, such as transforming a long hairstyle with bangs to a pixie cut, which requires removing the existing hair and inferring how the forehead would look, or transferring partially visible hair from a hat-wearing person in a different pose. Past solutions leverage StyleGAN for hallucinating any missing parts and producing a seamless face-hair composite through so-called GAN inversion or projection. However, there remains a challenge in controlling the hallucinations to accurately transfer hairstyle and preserve the face shape and identity of the input. To overcome this, we propose a multi-view optimization framework that uses two different views of reference composites to semantically guide occluded or ambiguous regions. Our optimization shares information between two poses, which allows us to produce high fidelity and realistic results from incomplete references. Our framework produces high-quality results and outperforms prior work in a user study that consists of significantly more challenging hair transfer scenarios than previously studied. Project page: https://stylegan-salon.github.io/.

Denoising diffusion models have shown great promise in human motion synthesis conditioned on natural language descriptions. However, integrating spatial constraints, such as pre-defined motion trajectories and obstacles, remains a challenge despite being essential for bridging the gap between isolated human motion and its surrounding environment. To address this issue, we propose Guided Motion Diffusion (GMD), a method that incorporates spatial constraints into the motion generation process. Specifically, we propose an effective feature projection scheme that manipulates motion representation to enhance the coherency between spatial information and local poses. Together with a new imputation formulation, the generated motion can reliably conform to spatial constraints such as global motion trajectories. Furthermore, given sparse spatial constraints (e.g. sparse keyframes), we introduce a new dense guidance approach to turn a sparse signal, which is susceptible to being ignored during the reverse steps, into denser signals to guide the generated motion to the given constraints. Our extensive experiments justify the development of GMD, which achieves a significant improvement over state-of-the-art methods in text-based motion generation while allowing control of the synthesized motions with spatial constraints.

We present a novel approach to single-view face relighting in the wild. Handling non-diffuse effects, such as global illumination or cast shadows, has long been a challenge in face relighting. Prior work often assumes Lambertian surfaces, simplified lighting models or involves estimating 3D shape, albedo, or a shadow map. This estimation, however, is error-prone and requires many training examples with lighting ground truth to generalize well. Our work bypasses the need for accurate estimation of intrinsic components and can be trained solely on 2D images without any light stage data, multi-view images, or lighting ground truth. Our key idea is to leverage a conditional diffusion implicit model (DDIM) for decoding a disentangled light encoding along with other encodings related to 3D shape and facial identity inferred from off-the-shelf estimators. We also propose a novel conditioning technique that eases the modeling of the complex interaction between light and geometry by using a rendered shading reference to spatially modulate the DDIM. We achieve state-of-the-art performance on standard benchmark Multi-PIE and can photorealistically relight in-the-wild images. Please visit our page: https://diffusion-face-relighting.github.io.

The joint visual-language model CLIP has enabled new and exciting applications, such as open-vocabulary segmentation, which can locate any segment given an arbitrary text query. In our research, we ask whether it is possible to discover semantic segments without any user guidance in the form of text queries or predefined classes, and label them using natural language automatically? We propose a novel problem zero-guidance segmentation and the first baseline that leverages two pre-trained generalist models, DINO and CLIP, to solve this problem without any fine-tuning or segmentation dataset. The general idea is to first segment an image into small over-segments, encode them into CLIP’s visual-language space, translate them into text labels, and merge semantically similar segments together. The key challenge, however, is how to encode a visual segment into a segment-specific embedding that balances global and local context information, both useful for recognition. Our main contribution is a novel attention-masking technique that balances the two contexts by analyzing the attention layers inside CLIP. We also introduce several metrics for the evaluation of this new task. With CLIP’s innate knowledge, our method can precisely locate the Mona Lisa painting among a museum crowd (Figure 1). More results are available at https://zero-guide-seg.github.io/.

Deep metric learning is a supervised learning paradigm to construct a meaningful vector space to represent complex objects. A successful application of deep metric learning to pointsets means that we can avoid expensive retrieval operations on objects such as documents and can significantly facilitate many machine learning and data mining tasks involving pointsets. We propose a self-supervised deep metric learning solution for pointsets. The novelty of our proposed solution lies in a self-supervision mechanism that makes use of a distribution distance for set ranking called the Earth's Mover Distance (EMD) to generate pseudo labels and a pointset augmentation method for supporting the learning solution. Our experimental studies on documents, graphs, and point clouds datasets show that our proposed solutions outperform baselines and state-of-the-art approaches under the unsupervised settings. The learned self-supervised representation can also be used as a pre-trained model, which can boost downstream tasks with a fine-tuning step and outperform state-of-the-art language models.

We present NeX, a new approach to novel view synthesis based on enhancements of multiplane images (MPI) that can reproduce view-dependent effects in real time. Unlike traditional MPI, our technique parameterizes each pixel as a linear combination of spherical basis functions learned from a neural network to model view-dependent effects and uses a hybrid implicit-explicit modeling strategy to improve fine detail. Moreover, we also present an extension to NeX, which leveragesknowledge distillationto train multiple MPIs for unbounded 360$^\circ$scenes. Our method is evaluated on several benchmark datasets: NeRF-Synthetic dataset, Light Field dataset, Real Forward-Facing dataset, Space dataset, as well asShiny, our new dataset that contains significantly more challenging view-dependent effects, such as the rainbow reflections on the CD. Our method outperforms other real-time rendering approaches on PSNR, SSIM, and LPIPS and can renderunbounded360$^\circ$scenes in real time.

While GANs have shown success in realistic image generation, the idea of using GANs for other tasks unrelated to synthesis is underexplored. Do GANs learn meaningful structural parts of objects during their attempt to reproduce those objects? And can image synthesis serve as an "upstream" representation learning task? In this work, we test these hypotheses and propose a simple and effective approach based on GANs for fundamental vision tasks: semantic part segmentation and landmark detection. With our approach, these tasks only require as few as one labeled example along with an unlabeled dataset, rather than thousands of examples. Our key idea is to leverage a trained GAN to extract a pixel-wise representation from the input image and use it as feature vectors for a segmentation network. Our experiments demonstrate that this GAN-derived representation is "readily discriminative" and produces surprisingly good results that are comparable to those from supervised baselines trained with significantly more labels. We believe this novel repurposing of GANs underlies a new class of unsupervised representation learning, which can generalize to many other tasks. More results are available at https://RepurposeGANs.github.io/.

Diffusion probabilistic models (DPMs) have achieved remarkable quality in image generation that rivals GANs'. But unlike GANs, DPMs use a set of latent variables that lack semantic meaning and cannot serve as a useful representation for other tasks. This paper explores the possibility of using DPMs for representation learning and seeks to extract a meaningful and decodable representation of an input image via autoencoding. Our key idea is to use a learnable encoder for discovering the high-level semantics, and a DPM as the decoder for modeling the remaining stochastic variations. Our method can encode any image into a two-part latent code where the first part is semantically meaningful and linear, and the second part captures stochastic details, allowing near-exact reconstruction. This capability enables challenging applications that currently foil GAN-based methods, such as attribute manipulation on real images. We also show that this two-level encoding improves denoising efficiency and naturally facilitates various downstream tasks including few-shot conditional sampling. Please visit our page: https://Diff-AE.github.io/

We propose a deep learning-based approach to localizing a small moving object with a single millimeter-wave frequency-modulated continuous-wave (FMCW) radar. The main challenge that foils conventional localization techniques, such as 3-D fast Fourier transform (3-D-FFT), Pisarenko method, multiple signal classification (MUSIC), estimation of signal parameters via rotational invariance technique (ESPRIT), Capon’s method, and Burg’s method, is the low signal-to-noise ratio of the reflected signal from millimeter-sized objects. Our key idea is to combine useful but noisy features from classical transforms [e.g., fast Fourier transform (FFT)] with neural networks that can refine and interpret those features into range and angle estimates by training on a large dataset of examples. Importantly, our networks were designed to be translation-equivariant, which enables accurate predictions of unseen object locations and improves the range and azimuth root mean square error (RMSE) scores by 34%–46% and 41%–60%, respectively, over state-of-the-art approaches. This pilot study establishes a new baseline for small-object tracking using FMCW and can enable tracking of small animals, such as ants inside the colony for behavior studies. Our first FMCW-small-object dataset and the source code are publicly available onhttps://github.com/shikuzen/RA-CNN.

While GANs have shown success in realistic image generation, the idea of using GANs for other tasks unrelated to synthesis is underexplored. Do GANs learn meaningful structural parts of objects during their attempt to reproduce those objects? In this work, we test this hypothesis and propose a simple and effective approach based on GANs for semantic part segmentation that requires as few as one label example along with an unlabeled dataset. Our key idea is to leverage a trained GAN to extract a pixel-wise representation from the input image and use it as feature vectors for a segmentation network. Our experiments demonstrate that this GAN-derived representation is "readily discriminative" and produces surprisingly good results that are comparable to those from supervised baselines trained with significantly more labels. We believe this novel repurposing of GANs underlies a new class of unsupervised representation learning, which can generalize to many other tasks. More results are available at https://RepurposeGANs.github.io/.

We present NeX, a new approach to novel view synthesis based on enhancements of multiplane image (MPI) that can reproduce next-level view-dependent effects—in real time. Unlike traditional MPI that uses a set of simple RGBα planes, our technique models view-dependent effects by instead parameterizing each pixel as a linear combination of basis functions learned from a neural network. Moreover, we propose a hybrid implicit-explicit modeling strategy that improves upon fine detail and produces state-of-the-art results. Our method is evaluated on benchmark forwardfacing datasets as well as our newly-introduced dataset designed to test the limit of view-dependent modeling with significantly more challenging effects such as the rainbow reflections on a CD. Our method achieves the best overall scores across all major metrics on these datasets with more than 1000× faster rendering time than the state of the art. For real-time demos, visit https://nex-mpi.github.io/

Deep metric learning is a supervised learning paradigm to construct a meaningful vector space to represent complex objects. A successful application of deep metric learning to pointsets means that we can avoid expensive retrieval operations on objects such as documents and can significantly facilitate many machine learning and data mining tasks involving pointsets. We propose a self-supervised deep metric learning solution for pointsets. The novelty of our proposed solution lies in a self-supervision mechanism that makes use of a distribution distance for set ranking called the Earth's Mover Distance (EMD) to generate pseudo labels. Our experimental studies on four documents datasets show that our proposed solutions outperform baselines and state-of-the-art approaches on unsupervised deep metric learning in most settings.

Recognizing movements during sleep is crucial for the monitoring of patients with sleep disorders, and the utilization of ultra-wideband (UWB) radar for the classification of human sleep postures has not been explored widely. This study investigates the performance of an off-the-shelf single antenna UWB in a novel application of sleep postural transition (SPT) recognition. The proposed Multi-View Learning, entitled SleepPoseNet or SPN, with time series data augmentation aims to classify four standard SPTs. SPN exhibits an ability to capture both time and frequency features, including the movement and direction of sleeping positions. The data recorded from 38 volunteers displayed that SPN with a mean accuracy of 73.7 ±0.8 % significantly outperformed the mean accuracy of 59.9 ±0.7 % obtained from deep convolution neural network (DCNN) in recent state-of-the-art work on human activity recognition using UWB. Apart from UWB system, SPN with the data augmentation can ultimately be adopted to learn and classify time series data in various applications.

This paper presents KeypointNet, an end-to-end geometric reasoning framework to learn an optimal set of category-specific keypoints, along with their detectors to predict 3D keypoints in a single 2D input image. We demonstrate this framework on 3D pose estimation task by proposing a differentiable pose objective that seeks the optimal set of keypoints for recovering the relative pose between two views of an object. Our network automatically discovers a consistent set of keypoints across viewpoints of a single object as well as across all object instances of a given object class. Importantly, we find that our end-to-end approach using no ground-truth keypoint annotations outperforms a fully supervised baseline using the same neural network architecture for the pose estimation task. The discovered 3D keypoints across the car, chair, and plane categories of ShapeNet are visualized at https://keypoints.github.io/

Given audio of President Barack Obama, we synthesize a high quality video of him speaking with accurate lip sync, composited into a target video clip. Trained on many hours of his weekly address footage, a recurrent neural network learns the mapping from raw audio features to mouth shapes. Given the mouth shape at each time instant, we synthesize high quality mouth texture, and composite it with proper 3D pose matching to change what he appears to be saying in a target video to match the input audio track. Our approach produces photorealistic results.

While prior depth from focus and defocus techniques operated on laboratory scenes, we introduce the first depth from focus (DfF) method capable of handling images from mobile phones and other hand-held cameras. Achieving this goal requires solving a novel uncalibrated DfF problem and aligning the frames to account for scene parallax. Our approach is demonstrated on a range of challenging cases and produces high quality results.

We reconstruct a controllable model of a person from a large photo collection that captures his or her persona, i.e., physical appearance and behavior. The ability to operate on unstructured photo collections enables modeling a huge number of people, including celebrities and other well photographed people without requiring them to be scanned. Moreover, we show the ability to drive or puppeteer the captured person B using any other video of a different person A. In this scenario, B acts out the role of person A, but retains his/her own personality and character. Our system is based on a novel combination of 3D face reconstruction, tracking, alignment, and multi-texture modeling, applied to the puppeteering problem. We demonstrate convincing results on a large variety of celebrities derived from Internet imagery and video.

We present an approach that takes a single photograph of a child as input and automatically produces a series of age-progressed outputs between 1 and 80 years of age, accounting for pose, expression, and illumination. Leveraging thousands of photos of children and adults at many ages from the Internet, we first show how to compute average image subspaces that are pixel-to-pixel aligned and model variable lighting. These averages depict a prototype man and woman aging from 0 to 80, under any desired illumination, and capture the differences in shape and texture between ages. Applying these differences to a new photo yields an age progressed result. Contributions include relightable age subspaces, a novel technique for subspace-to-subspace alignment, and the most extensive evaluation of age progression techniques in the literature.