Abstract Single‐image 3D reconstruction with large reconstruction models (LRMs) has advanced rapidly, yet reconstructions often exhibit geometric inconsistencies and misaligned details that limit fidelity. We introduce GeoFusionLRM, a geometry‐aware self‐correction framework that leverages the model's own normal and depth predictions to refine structural accuracy. Unlike prior approaches that rely solely on features extracted from the input image, GeoFusionLRM feeds back geometric cues through a dedicated transformer and fusion module, enabling the model to correct errors and enforce consistency with the conditioning image. This design improves the alignment between the reconstructed mesh and the input views without additional supervision or external signals. Extensive experiments demonstrate that GeoFusionLRM achieves sharper geometry, more consistent normals, and higher fidelity than state‐of‐the‐art LRM baselines.

Semantic image editing involves inpainting pixels guided by a semantic map. This is a challenging task, as the inpainted regions must both align harmoniously with the surrounding context and strictly adhere to the semantic constraints. Most prior methods approach this by attempting to encode all necessary information from the erased regions alone. However, when adding new objects-such as a car-to a scene, their style often cannot be inferred solely from the surrounding context. On the other hand, the models that can output diverse generations struggle to output images that have seamless boundaries between the generated and unerased parts. In this work, we propose a framework that can encode visible and partially visible objects with a novel mechanism to achieve consistency in the style encoding and final generations. We extensively compare with previous conditional image generation and semantic image editing algorithms. Our extensive experiments show that our method significantly improves over the state-of-the-art. Our method not only achieves better quantitative results but also provides diverse results. Demo and code will be released.

Generative Adversarial Networks (GANs) have emerged as powerful tools for high-quality image generation and real image editing by manipulating their latent spaces. Recent advancements in GANs include 3D-aware models such as EG3D, which feature efficient triplane-based architectures capable of reconstructing 3D geometry from single images. However, limited attention has been given to providing an integrated framework for 3D-aware, high-quality, reference-based image editing. This study addresses this gap by exploring and demonstrating the effectiveness of the triplane space for advanced reference-based edits. Our novel approach integrates encoding, automatic localization, spatial disentanglement of triplane features, and fusion learning to achieve the desired edits. We demonstrate how our approach excels across diverse domains, including human faces, 360-degree heads, animal faces, partially stylized edits like cartoon faces, full-body clothing edits, and edits on class-agnostic samples. Our method shows stateof-the-art performance over relevant latent direction, text, and image-guided 2D and 3D-aware diffusion and GAN methods, both qualitatively and quantitatively.1

3D head stylization transforms realistic facial features into artistic representations, enhancing user engagement across gaming and virtual reality applications. While 3D-aware generators have made significant advancements, many 3D stylization methods primarily provide near-frontal views and struggle to preserve the unique identities of original subjects, often resulting in outputs that lack diversity and individuality. This paper addresses these challenges by leveraging the PanoHead model, synthesizing images from a comprehensive 360-degree perspective. We propose a novel framework that employs negative log-likelihood distillation (LD) to enhance identity preservation and improve stylization quality. By integrating multi-view grid score and mirror gradients within the 3D GAN architecture and introducing a score rank weighing technique, our approach achieves substantial qualitative and quantitative improvements. Our findings not only advance the state of 3D head stylization but also provide valuable insights into effective distillation processes between diffusion models and GANs, focusing on the critical issue of identity preservation. Please visit the https://three-bee.github.io/head_stylization for more visuals.

We propose RoPECraft, a training-free video motion transfer method for diffusion transformers that operates solely by modifying their rotary positional embeddings (RoPE). We first extract dense optical flow from a reference video, and utilize the resulting motion offsets to warp the complex-exponential tensors of RoPE, effectively encoding motion into the generation process. These embeddings are then further optimized during denoising time steps via trajectory alignment between the predicted and target velocities using a flow-matching objective. To keep the output faithful to the text prompt and prevent duplicate generations, we incorporate a regularization term based on the phase components of the reference video’s Fourier transform, projecting the phase angles onto a smooth manifold to suppress high-frequency artifacts. Experiments on benchmarks reveal that RoPECraft outperforms all recently published methods, both qualitatively and quantitatively.

3D GAN inversion aims to project a single image into the latent space of a 3D Generative Adversarial Network (GAN), thereby achieving 3D geometry reconstruction. While there exist encoders that achieve good results in 3D GAN inversion, they are predominantly built on EG3D, which specializes in synthesizing near-frontal views and is limiting in synthesizing comprehensive 3D scenes from diverse viewpoints. In contrast to existing approaches, we propose a novel framework built on PanoHead, which excels in synthesizing images from a 360-degree perspective. To achieve realistic 3D modeling of the input image, we introduce a dual encoder system tailored for high-fidelity reconstruction and realistic generation from different viewpoints. Accompanying this, we propose a stitching framework on the triplane domain to get the best predictions from both. To achieve seamless stitching, both encoders must output consistent results despite being specialized for different tasks. For this reason, we carefully train these encoders using specialized losses, including an adversarial loss based on our novel occlusion-aware triplane discriminator. Experiments reveal that our approach surpasses the existing encoder training methods qualitatively and quantitatively.

Advanced image editing techniques, particularly inpainting, are essential for seamlessly removing unwanted elements while preserving visual integrity. Traditional GAN-based methods have achieved notable success, but recent advancements in diffusion models have produced superior results due to their training on large-scale datasets, enabling the generation of remarkably realistic inpainted images. Despite their strengths, diffusion models often struggle with object removal tasks without explicit guidance, leading to unintended hallucinations of the removed object. To address this issue, we introduce CLIPAway, a novel approach leveraging CLIP embeddings to focus on background regions while excluding foreground elements. CLIPAway enhances inpainting accuracy and quality by identifying embeddings that prioritize the background, thus achieving seamless object removal. Unlike other methods that rely on specialized training datasets or costly manual annotations, CLIPAway provides a flexible, plug-and-play solution compatible with various diffusion-based inpainting techniques.

Estimating 3D human texture from a single image is essential in graphics and vision. It requires learning a mapping function from input images of humans with diverse poses into the parametric (uv) space and reasonably hallucinating invisible parts. To achieve a high-quality 3D human texture estimation, we propose a framework that adaptively samples the input by a deformable convolution where offsets are learned via a deep neural network. Additionally, we describe a novel cycle consistency loss that improves view generalization. We further propose to train our framework with an uncertainty-based pixel-level image reconstruction loss, which enhances color fidelity. We compare our method against the state-of-the-art approaches and show significant qualitative and quantitative improvements.

This paper presents a method to reconstruct high-quality textured 3D models from single images. Current methods rely on datasets with expensive annotations; multi-view images and their camera parameters. Our method relies on GAN generated multi-view image datasets which have a negligible annotation cost. However, they are not strictly multi-view consistent and sometimes GANs output distorted images. This results in degraded reconstruction qualities. In this work, to overcome these limitations of generated datasets, we have two main contributions which lead us to achieve state-of-the-art results on challenging objects: 1) A robust multi-stage learning scheme that gradually relies more on the models own predictions when calculating losses and 2) A novel adversarial learning pipeline with online pseudo-ground truth generations to achieve fine details. Our work provides a bridge from 2D supervisions of GAN models to 3D reconstruction models and removes the expensive annotation efforts. We show significant improvements over previous methods whether they were trained on GAN generated multi-view images or on real images with expensive annotations.

This article presents a comprehensive evaluation of instance segmentation models with respect to real-world image corruptions as well as out-of-domain image collections, e.g., images captured by a different set-up than the training dataset. The out-of-domain image evaluation shows the generalization capability of models, an essential aspect of real-world applications, and an extensively studied topic of domain adaptation. These presented robustness and generalization evaluations are important when designing instance segmentation models for real-world applications and picking an off-the-shelf pretrained model to directly use for the task at hand. Specifically, this benchmark study includes state-of-the-art network architectures, network backbones, normalization layers, models trained starting from scratch versus pretrained networks, and the effect of multitask training on robustness and generalization. Through this study, we gain several insights. For example, we find that group normalization (GN) enhances the robustness of networks across corruptions where the image contents stay the same but corruptions are added on top. On the other hand, batch normalization (BN) improves the generalization of the models across different datasets where statistics of image features change. We also find that single-stage detectors do not generalize well to larger image resolutions than their training size. On the other hand, multistage detectors can easily be used on images of different sizes. We hope that our comprehensive study will motivate the development of more robust and reliable instance segmentation models.

We present a novel image inversion framework and a training pipeline to achieve high-fidelity image inversion with high-quality attribute editing. Inverting real images into StyleGAN's latent space is an extensively studied problem, yet the trade-off between the image reconstruction fidelity and image editing quality remains an open challenge. The low-rate latent spaces are limited in their expressiveness power for high-fidelity reconstruction. On the other hand, high-rate latent spaces result in degradation in editing quality. In this work, to achieve high-fidelity inversion, we learn residual features in higher latent codes that lower latent codes were not able to encode. This enables preserving image details in reconstruction. To achieve high-quality editing, we learn how to transform the residual features for adapting to manipulations in latent codes. We train the framework to extract residual features and transform them via a novel architecture pipeline and cycle consistency losses. We run extensive experiments and compare our method with state-of-the-art inversion methods. Qualitative metrics and visual comparisons show significant improvements. Code: https://github.com/hamzapehlivanIStyleRes.

Recent inversion methods have shown that real images can be inverted into StyleGAN’s latent space and numerous edits can be achieved on those images thanks to the semantically rich feature representations of well-trained GAN models. However, extensive research has also shown that image inversion is challenging due to the trade-off between high-fidelity reconstruction and editability. In this paper, we tackle an even more difficult task, inverting erased images into GAN’s latent space for realistic inpaintings and editings. Furthermore, by augmenting inverted latent codes with different latent samples, we achieve diverse inpaintings. Specifically, we propose to learn an encoder and mixing network to combine encoded features from erased images with StyleGAN’s mapped features from random samples. To encourage the mixing network to utilize both inputs, we train the networks with generated data via a novel set-up. We also utilize higher-rate features to prevent color inconsistencies between the inpainted and unerased parts. We run extensive experiments and compare our method with state-of-the-art inversion and inpainting methods. Qualitative metrics and visual comparisons show significant improvements.

We propose an image-to-image translation framework for facial attribute editing with disentangled interpretable latent directions. Facial attribute editing task faces the challenges of targeted attribute editing with controllable strength and disentanglement in the representations of attributes to preserve the other attributes during edits. For this goal, inspired by the latent space factorization works of fixed pretrained GANs, we design the attribute editing by latent space factorization, and for each attribute, we learn a linear direction that is orthogonal to the others. We train these directions with orthogonality constraints and disentanglement losses. To project images to semantically organized latent spaces, we set an encoder-decoder architecture with attention-based skip connections. We extensively compare with previous image translation algorithms and editing with pretrained GAN works. Our extensive experiments show that our method significantly improves over the state-of-the-arts.

This paper presents a method to achieve fine detailed texture learning for 3D models that are reconstructed from both multi-view and single-view images. The framework is posed as an adaptation problem and is done progressively where in the first stage, we focus on learning accurate geometry, whereas in the second stage, we focus on learning the texture with a generative adversarial network. The contributions of the paper are in the generative learning pipeline where we propose two improvements. First, since the learned textures should be spatially aligned, we propose an attention mechanism that relies on the learnable positions of pixels. Second, since discriminator receives aligned texture maps, we augment its input with a learnable embedding which improves the feedback to the generator. We achieve significant improvements on multi-view sequences from Tripod dataset as well as on single-view image datasets, Pascal 3D+ and CUB. We demonstrate that our method achieves superior 3D textured models compared to the previous works.

Partial convolution weights convolutions with binary masks and renormalizes on valid pixels. It was originally proposed for image inpainting task because a corrupted image processed by a standard convolutional often leads to artifacts. Therefore, binary masks are constructed that define the valid and corrupted pixels, so that partial convolution results are only calculated based on valid pixels. It has been also used for conditional image synthesis task, so that when a scene is generated, convolution results of an instance depend only on the feature values that belong to the same instance. One of the unexplored applications for partial convolution is padding which is a critical component of modern convolutional networks. Common padding schemes make strong assumptions about how the padded data should be extrapolated. We show that these padding schemes impair model accuracy, whereas partial convolution based padding provides consistent improvements across a range of tasks. In this article, we review partial convolution applications under one framework. We conduct a comprehensive study of the partial convolution based padding on a variety of computer vision tasks, including image classification, 3D-convolution-based action recognition, and semantic segmentation. Our results suggest that partial convolution-based padding shows promising improvements over strong baselines.

We analyze why the orthogonality penalty improves quantization in deep neural networks. Using results from perturbation theory as well as through extensive experiments with Resnet50, Resnet101, and VGG19 models, we mathematically and experimentally show that improved quantization accuracy resulting from orthogonality constraint stems primarily from reduced condition numbers, which is the ratio of largest to smallest singular values of weight matrices, more so than reduced spectral norms, in contrast to the explanations in previous literature. We also show that the orthogonality penalty improves quantization even in the presence of a state-of-the-art quantized retraining method. Our results show that, when the orthogonality penalty is used with quantized retraining, ImageNet Top5 accuracy loss from 4- to 8-bit quantization is reduced by up to 7% for Resnet50, and up to 10% for Resnet101, compared to quantized retraining with no orthogonality penalty.

Unsupervised landmark learning is the task of learning semantic keypoint-like representations without the use of expensive input keypoint annotations. A popular approach is to factorize an image into a pose and appearance data stream, then to reconstruct the image from the factorized components. The pose representation should capture a set of consistent and tightly localized landmarks in order to facilitate reconstruction of the input image. Ultimately, we wish for our learned landmarks to focus on the foreground object of interest. However, the reconstruction task of the entire image forces the model to allocate landmarks to model the background. Using a motion-based foreground assumption, this work explores the effects of factorizing the reconstruction task into separate foreground and background reconstructions in an unsupervised way, allowing the model to condition only the foreground reconstruction on the unsupervised landmarks. Our experiments demonstrate that the proposed factorization results in landmarks that are focused on the foreground object of interest when measured against ground-truth foreground masks. Furthermore, the rendered background quality is also improved as ill-suited landmarks are no longer forced to model this content. We demonstrate this improvement via improved image fidelity in a video-prediction task. Code is available at https://github.com/NVIDIA/UnsupervisedLandmarkLearning.

Humans can easily infer the underlying 3D geometry and texture of an object only from a single 2D image. Current computer vision methods can do this, too, but suffer from view generalization problems – the models inferred tend to make poor predictions of appearance in novel views. As for generalization problems in machine learning, the difficulty is balancing single-view accuracy (cf. training error; bias) with novel view accuracy (cf. test error; variance). We describe a class of models whose geometric rigidity is easily controlled to manage this tradeoff. We describe a cycle consistency loss that improves view generalization (roughly, a model from a generated view should predict the original view well). View generalization of textures requires that models share texture information, so a car seen from the back still has headlights because other cars have headlights. We describe a cycle consistency loss that encourages model textures to be aligned, so as to encourage sharing. We compare our method against the state-of-the-art method and show both qualitative and quantitative improvements.

Generative Adversarial Networks (GANs) produce impressive results on unconditional image generation when powered with large-scale image datasets. Yet generated images are still easy to spot especially on datasets with high variance (e.g. bedroom, church). In this paper, we propose various improvements to further push the boundaries in image generation. Specifically, we propose a novel dual contrastive loss and show that, with this loss, discriminator learns more generalized and distinguishable representations to incentivize generation. In addition, we revisit attention and extensively experiment with different attention blocks in the generator. We find attention to be still an important module for successful image generation even though it was not used in the recent state-of-the-art models. Lastly, we study different attention architectures in the discriminator, and propose a reference attention mechanism. By combining the strengths of these remedies, we improve the compelling state-of-the-art Fréchet Inception Distance (FID) by at least 17.5% on several benchmark datasets. We obtain even more significant improvements on compositional synthetic scenes (up to 47.5% in FID).

Generating computer graphics (CG) rendered synthetic images has been widely used to create simulation environments for robotics/autonomous driving and generate labeled data. Yet, the problem of training models purely with synthetic data remains challenging due to the considerable domain gaps caused by current limitations on rendering. In this paper, we propose a simple yet effective domain adaptation framework towards closing such gap at image level. Unlike many GAN-based approaches, our method aims to match the covariance of the universal feature embeddings across domains, making the adaptation a fast, convenient step and avoiding the need for potentially difficult GAN training. To align domains more precisely, we further propose a conditional covariance matching framework which iteratively estimates semantic segmentation regions and conditionally matches the class-wise feature covariance given the segmentation regions. We demonstrate that both tasks can mutually refine and considerably improve each other, leading to state-of-the-art domain adaptation results. Extensive experiments under multiple synthetic-to-real settings show that our approach exceeds the performance of latest domain adaptation approaches. In addition, we offer a quantitative analysis where our framework shows considerable reduction in Frechet Inception distance between source and target domains, demonstrating the effectiveness of this work in bridging the synthetic-to-real domain gap.

Conditional image synthesis for generating photorealistic images serves various applications for content editing to content generation. Previous conditional image synthesis algorithms mostly rely on semantic maps, and often fail in complex environments where multiple instances occlude each other. We propose a panoptic aware image synthesis network to generate high fidelity and photorealistic images conditioned on panoptic maps which unify semantic and instance information. To achieve this, we efficiently use panoptic maps in convolution and upsampling layers. We show that with the proposed changes to the generator, we can improve on the previous state-of-the-art methods by generating images in complex instance interaction environments in higher fidelity and tiny objects in more details. Furthermore, our proposed method also outperforms the previous state-of-the-art methods in metrics of mean IoU (Intersection over Union), and detAP (Detection Average Precision).

Synthesizing larger texture images from a smaller exemplar is an important task in graphics and vision. The conventional CNNs, recently adopted for synthesis, require to train and test on the same set of images and fail to generalize to unseen images. This is mainly because those CNNs fully rely on convolutional and upsampling layers that operate locally and not suitable for a task as global as texture synthesis. In this work, inspired by the repetitive nature of texture patterns, we find that texture synthesis can be viewed as (local) \textit{upsampling} in the Fast Fourier Transform (FFT) domain. However, FFT of natural images exhibits high dynamic range and lacks local correlations. Therefore, to train CNNs we design a framework to perform FFT upsampling in feature space using deformable convolutions. Such design allows our framework to generalize to unseen images, and synthesize textures in a single pass. Extensive evaluations confirm that our method achieves state-of-the-art performance both quantitatively and qualitatively.

Learning to synthesize high frame rate videos via interpolation requires large quantities of high frame rate training videos, which, however, are scarce, especially at high resolutions. Here, we propose unsupervised techniques to synthesize high frame rate videos directly from low frame rate videos using cycle consistency. For a triplet of consecutive frames, we optimize models to minimize the discrepancy between the center frame and its cycle reconstruction, obtained by interpolating back from interpolated intermediate frames. This simple unsupervised constraint alone achieves results comparable with supervision using the ground truth intermediate frames. We further introduce a pseudo supervised loss term that enforces the interpolated frames to be consistent with predictions of a pre-trained interpolation model. The pseudo supervised loss term, used together with cycle consistency, can effectively adapt a pre-trained model to a new target domain. With no additional data and in a completely unsupervised fashion, our techniques significantly improve pre-trained models on new target domains, increasing PSNR values from 32.84dB to 33.05dB on the Slowflow and from 31.82dB to 32.53dB on the Sintel evaluation datasets.

Deep convolutional neural networks (DCNNs) have become a very powerful tool in visual perception. DCNNs have applications in autonomous robots, security systems, mobile phones, and automobiles, where high throughput of the feedforward evaluation phase and power efficiency are important. Because of this increased usage, many field-programmable gate array (FPGA)-based accelerators have been proposed. In this paper, we present an optimized streaming method for DCNNs' hardware accelerator on an embedded platform. The streaming method acts as a compiler, transforming a high-level representation of DCNNs into operation codes to execute applications in a hardware accelerator. The proposed method utilizes maximum computational resources available based on a novel-scheduled routing topology that combines data reuse and data concatenation. It is tested with a hardware accelerator implemented on the Xilinx Kintex-7 XC7K325T FPGA. The system fully explores weight-level and node-level parallelizations of DCNNs and achieves a peak performance of 247 G-ops while consuming less than 4 W of power. We test our system with applications on object classification and object detection in real-world scenarios. Our results indicate high-performance efficiency, outperforming all other presented platforms while running these applications.