In this work we develop a generalizable and efficient Neural Radiance Field (NeRF) pipeline for high-fidelity free-viewpoint human body synthesis under settings with sparse camera views. Though existing NeRF-based methods can synthesize rather realistic details for human body, they tend to produce poor results when the input has self-occlusion, especially for unseen humans under sparse views. Moreover, these methods often require a large number of sampling points for rendering, which leads to low efficiency and limits their real-world applicability. To address these challenges, we propose a Geometry-guided Progressive NeRF (GP-NeRF). In particular, to better tackle self-occlusion, we devise a geometry-guided multi-view feature integration approach that utilizes the estimated geometry prior to integrate the incomplete information from input views and construct a complete geometry volume for the target human body. Meanwhile, for achieving higher rendering efficiency, we introduce a geometry-guided progressive rendering pipeline, which leverages the geometric feature volume and the predicted density values to progressively reduce the number of sampling points and speed up the rendering process. Experiments on the ZJU-MoCap and THUman datasets show that our method outperforms the state-of-the-arts significantly across multiple generalization settings, while the time cost is reduced via >70% via applying our efficient progressive rendering pipeline.
论文检索
输入标题、作者或关键词，从 1,645 篇学术成果中精准定位
Today’s generative models are capable of synthesizing high-fidelity images, but each model specializes on a specific target domain. This raises the need for model merging: combining two or more pretrained generative models into a single unified one. In this work we tackle the problem of model merging, given two constraints that often come up in the real world: (1) no access to the original training data, and (2) without increasing the size of the neural network. To the best of our knowledge, model merging under these constraints has not been studied thus far. We propose a novel, two-stage solution. In the first stage, we transform the weights of all the models to the same parameter space by a technique we term model rooting. In the second stage, we merge the rooted models by averaging their weights and fine-tuning them for each specific domain, using only data generated by the original trained models. We demonstrate that our approach is superior to baseline methods and to existing transfer learning techniques, and investigate several applications.
We present ChunkyGAN-a novel paradigm for modeling and editing images using generative adversarial networks. Unlike previous techniques seeking a global latent representation of the input image, our approach subdivides the input image into a set of smaller components (chunks) specified either manually or automatically using a pre-trained segmentation network. For each chunk, the latent code of a generative network is estimated locally with greater accuracy thanks to a smaller number of constraints. Moreover, during the optimization of latent codes, segmentation can further be refined to improve matching quality. This process enables high-quality projection of the original image with spatial disentanglement that previous methods would find challenging to achieve. To demonstrate the advantage of our approach, we evaluated it quantitatively and also qualitatively in various image editing scenarios that benefit from the higher reconstruction quality and local nature of the approach. Our method is flexible enough to manipulate even out-of-domain images that would be hard to reconstruct using global techniques.
Unleashing Transformers: Parallel Token Prediction with Discrete Absorbing Diffusion for Fast High-Resolution Image Generation from Vector-Quantized Codes
PDF ↗Whilst diffusion probabilistic models can generate high quality image content, key limitations remain in terms of both generating high-resolution imagery and their associated high computational requirements. Recent Vector-Quantized image models have overcome this limitation of image resolution but are prohibitively slow and unidirectional as they generate tokens via element-wise autoregressive sampling from the prior. By contrast, in this paper we propose a novel discrete diffusion probabilistic model prior which enables parallel prediction of Vector-Quantized tokens by using an unconstrained Transformer architecture as the backbone. During training, tokens are randomly masked in an order-agnostic manner and the Transformer learns to predict the original tokens. This parallelism of Vector-Quantized token prediction in turn facilitates unconditional generation of globally consistent high-resolution and diverse imagery at a fraction of the computational expense. In this manner, we can generate image resolutions exceeding that of the original training set samples whilst additionally provisioning per-image likelihood estimates (in a departure from generative adversarial approaches). Our approach achieves state-of-the-art results in terms of the manifold overlap metrics Coverage (LSUN Bedroom: 0.83; LSUN Churches: 0.73; FFHQ: 0.80) and Density (LSUN Bedroom: 1.51; LSUN Churches: 1.12; FFHQ: 1.20), and performs competitively on FID (LSUN Bedroom: 3.64; LSUN Churches: 4.07; FFHQ: 6.11) whilst offering advantages in terms of both computation and reduced training set requirements.
Image outpainting, which is well studied with Convolution Neural Network (CNN) based framework, has recently drawn more attention in computer vision. However, CNNs rely on inherent inductive biases to achieve effective sample learning, which may degrade the performance ceiling. In this paper, motivated by the flexible self-attention mechanism with minimal inductive biases in transformer architecture, we reframe the generalised image outpainting problem as a patch-wise sequence-to-sequence autoregression problem, enabling query-based image outpainting. Specifically, we propose a novel hybrid vision-transformer-based encoder-decoder framework, named Query Outpainting TRansformer (QueryOTR), for extrapolating visual context all-side around a given image. Patch-wise mode’s global modeling capacity allows us to extrapolate images from the attention mechanism’s query standpoint. A novel Query Expansion Module (QEM) is designed to integrate information from the predicted queries based on the encoder’s output, hence accelerating the convergence of the pure transformer even with a relatively small dataset. To further enhance connectivity between each patch, the proposed Patch Smoothing Module (PSM) re-allocates and averages the overlapped regions, thus providing seamless predicted images. We experimentally show that QueryOTR could generate visually appealing results smoothly and realistically against the state-of-the-art image outpainting approaches.
Furnishing and rendering indoor scenes has been a long-standing task for interior design, where artists create a conceptual design for the space, build a 3D model of the space, decorate, and then perform rendering. Although the task is important, it is tedious and requires tremendous effort. In this paper, we introduce a new problem of domain-specific indoor scene image synthesis, namely neural scene decoration. Given a photograph of an empty indoor space and a list of decorations with layout determined by user, we aim to synthesize a new image of the same space with desired furnishing and decorations. Neural scene decoration can be applied to create conceptual interior designs in a simple yet effective manner. Our attempt to this research problem is a novel scene generation architecture that transforms an empty scene and an object layout into a realistic furnished scene photograph. We demonstrate the performance of our proposed method by comparing it with conditional image synthesis baselines built upon prevailing image translation approaches both qualitatively and quantitatively. We conduct extensive experiments to further validate the plausibility and aesthetics of our generated scenes.
We tackle the problem of object-centric learning on point clouds, which is crucial for high-level relational reasoning and scalable machine intelligence. In particular, we introduce a framework, SPAIR3D, to factorize a 3D point cloud into a spatial mixture model where each component corresponds to one object. To model the spatial mixture model on point clouds, we derive the Chamfer Mixture Loss, which fits naturally into our variational training pipeline. Moreover, we adopt an object-specification scheme that describes each object’s location relative to its local voxel grid cell. Such a scheme allows SPAIR3D to model scenes with an arbitrary number of objects. We evaluate our method on the task of unsupervised scene decomposition. Experimental results demonstrate that SPAIR3D has strong scalability and is capable of detecting and segmenting an unknown number of objects from a point cloud in an unsupervised manner.
Generative Adversarial Networks (GANs) have been widely applied in modeling diverse image distributions. However, despite its impressive applications, the structure of the latent space in GANs largely remains as a black-box, leaving its controllable generation an open problem, especially when spurious correlations between different semantic attributes exist in the image distributions. To address this problem, previous methods typically learn linear directions or individual channels that control semantic attributes in the image space. However, they often suffer from imperfect disentanglement, or are unable to obtain multi-directional controls. In this work, in light of the above challenges, we propose a novel approach that discovers nonlinear controls, which enables multi-directional manipulation as well as effective disentanglement, based on gradient information in the learned GAN latent space. More specifically, we first learn interpolation directions by following the gradients from classification networks trained separately on the attributes, and then navigate the latent space by exclusively controlling channels activated for the target attribute in the learned directions. Empirically, with small training data, our approach is able to gain fine-grained controls over a diverse set of bi-directional and multi-directional attributes, and we showcase its ability to achieve disentanglement significantly better than state-of-the-art methods both qualitatively and quantitatively.
TREND: Truncated Generalized Normal Density Estimation of Inception Embeddings for GAN Evaluation
PDF ↗Evaluating image generation models such as generative adversarial networks (GANs) is a challenging problem. A common approach is to compare the distributions of the set of ground truth images and the set of generated test images. The Frechet Inception distance is one of the most widely used metrics for evaluation of GANs, which assumes that the features from a trained Inception model for a set of images follow a normal distribution. In this paper, we argue that this is an over-simplified assumption, which may lead to unreliable evaluation results, and more accurate density estimation can be achieved using a truncated generalized normal distribution. Based on this, we propose a novel metric for accurate evaluation of GANs, named TREND (TRuncated gEneralized Normal Density estimation of inception embeddings). We demonstrate that our approach significantly reduces errors of density estimation, which consequently eliminates the risk of faulty evaluation results. Furthermore, we show that the proposed metric significantly improves robustness of evaluation results against variation of the number of image samples.
Non-autoregressive generative transformers recently demonstrated impressive image generation performance, and orders of magnitude faster sampling than their autoregressive counterparts. However, optimal parallel sampling from the true joint distribution of visual tokens remains an open challenge. In this paper we introduce Token-Critic, an auxiliary model to guide the sampling of a non-autoregressive generative transformer. Given a masked-and-reconstructed real image, the Token-Critic model is trained to distinguish which visual tokens belong to the original image and which were sampled by the generative transformer. During non-autoregressive iterative sampling, Token-Critic is used to select which tokens to accept and which to reject and resample. Coupled with Token-Critic, a state-of-the-art generative transformer significantly improves its performance, and outperforms recent diffusion models and GANs in terms of the trade-off between generated image quality and diversity, in the challenging class-conditional ImageNet generation.
Image Inpainting is an ill-posed problem since there are diverse possible counterparts for the missing areas. The challenge of inpainting is to keep the ""corrupted region"" content consistent with the background and generate a variety of reasonable texture details. However, existing one-stage methods that directly output the inpainting results have to make a trade-off between diversity and consistency. The two-stage methods as the current trend can circumvent such shortcomings. These methods predict diverse structural priors in the first stage and focus on rich texture details generation in the second stage. However, all two-stage methods require autoregressive models to predict the probability distribution of the structural priors, which significantly limits the inference speed. In addition, their discretization assumption of prior distribution reduces the diversity of the inpainting results. We propose Flow-Fill, a novel two-stage image inpainting framework that utilizes a conditional normalizing flow model to generate diverse structural priors in the first stage. Flow-Fill can directly estimate the joint probability density of the missing regions as a flow-based model without reasoning pixel by pixel. Hence it achieves real-time inference speed and eliminates discretization assumptions. In addition, as a reversible model, Flow-Fill can invert the latent variables for a specified region, which allows us to make the inference process as semantic image editing. Experiments on benchmark datasets validate that Flow-Fill achieves superior diversity and fidelity in image inpainting qualitatively and quantitatively.
Object reconstruction from 3D point clouds has achieved impressive progress in the computer vision and computer graphics research field. However, reconstruction from time-varying point clouds (a.k.a. 4D point clouds) is generally overlooked. In this paper, we propose a new network architecture, namely RFNet-4D, that jointly reconstruct objects and their motion flows from 4D point clouds. The key insight is that simultaneously performing both tasks via learning spatial and temporal features from a sequence of point clouds can leverage individual tasks, leading to improved overall performance. To prove this ability, we design a temporal vector field learning module using unsupervised learning approach for flow estimation, leveraged by supervised learning of spatial structures for object reconstruction. Extensive experiments and analyses on benchmark dataset validated the effectiveness and efficiency of our method. As shown in experimental results, our method achieves state-of-the-art performance on both flow estimation and object reconstruction while performing much faster than existing methods in both training and inference. Our code and data are available at https://github.com/hkust-vgd/RFNet-4D.
We present LiDARGen, a novel, effective, and controllable generative model that produces realistic LiDAR point cloud sensory readings. Our method leverages the powerful score-matching energy-based model and formulates the point cloud generation process as a stochastic denoising process in the equirectangular view. This model allows us to sample diverse and high-quality point cloud samples with guaranteed physical feasibility and controllability. We validate the effectiveness of our method on the challenging KITTI-360 and NuScenes datasets. The quantitative and qualitative results show that our approach produces more realistic samples than other generative models. Furthermore, LiDARGen can sample point clouds conditioned on inputs without retraining. We demonstrate that our proposed generative model could be directly used to densify LiDAR point clouds.
Score-based generative models (SGMs) have recently emerged as a promising class of generative models. However, a fundamental limitation is that their inference is very slow due to a need for many (e.g., 2000) iterations of sequential computations. An intuitive acceleration method is to reduce the sampling iterations which however causes severe performance degradation. We investigate this problem by viewing the diffusion sampling process as a Metropolis adjusted Langevin algorithm, which helps reveal the underlying cause to be ill-conditioned curvature. Under this insight, we propose a model-agnostic preconditioned diffusion sampling (PDS) method that leverages matrix preconditioning to alleviate the aforementioned problem. Crucially, PDS is proven theoretically to converge to the original target distribution of a SGM, no need for retraining. Extensive experiments on three image datasets with a variety of resolutions and diversity validate that PDS consistently accelerates off-the-shelf SGMs whilst maintaining the synthesis quality. In particular, PDS can accelerate by up to 29x on more challenging high resolution (1024x1024) image generation.
Denoising Diffusion Probabilistic Model (DDPM) is able to make flexible conditional image generation from prior noise to real data, by introducing an independent noise-aware classifier to provide conditional gradient guidance at each time step of denoising process. However, due to the ability of the classifier to easily discriminate an incompletely generated image only with high-level structure, the gradient, which is a kind of class information guidance, tends to vanish early, leading to the collapse from conditional generation process into the unconditional process. To address this problem, we propose two simple but effective approaches from two perspectives. For sampling procedure, we introduce the entropy of predicted distribution as the measure of guidance vanishing level and propose an entropy-aware scaling method to adaptively recover the conditional semantic guidance. For the training stage, we propose the entropy-aware optimization objectives to alleviate the overconfident prediction for noisy data. On ImageNet1000 256x256, with our proposed sampling scheme and trained classifier, the pretrained conditional and unconditional DDPM model can achieve 10.89% (4.59 to 4.09) and 43.5% (12 to 6.78) FID improvement, respectively. Code is available at https://github.com/ZGCTroy/ED-DPM.
Despite the rapid development of Neural Radiance Field (NeRF), the necessity of dense covers largely prohibits its wider applications. While several recent works have attempted to address this issue, they either operate with sparse views (yet still, a few of them) or on simple objects/scenes. In this work, we consider a more ambitious task: training neural radiance field, over realistically complex visual scenes, by “looking only once”, i.e., using only a single view. To attain this goal, we present a Single View NeRF (SinNeRF) framework consisting of thoughtfully designed semantic and geometry regularizations. Specifically, SinNeRF constructs a semi-supervised learning process, where we introduce and propagate geometry pseudo labels and semantic pseudo labels to guide the progressive training process. Extensive experiments are conducted on complex scene benchmarks, including NeRF synthetic dataset, Local Light Field Fusion dataset, and DTU dataset. We show that even without pre-training on multi-view datasets, SinNeRF can yield photo-realistic novel-view synthesis results. Under the single image setting, SinNeRF significantly outperforms the current state-of-the-art NeRF baselines in all cases. Project page: https://vita-group.github.io/SinNeRF/
In this paper, we propose a region adaptive makeup transfer GAN, called RamGAN, for precise region-level makeup transfer. Compared to face-level transfer methods, our RamGAN uses spatial-aware Region Attentive Morphing Module (RAMM) to encode Region Attentive Matrices (RAMs) for local regions like lips, eye shadow and skin. After that, the Region Style Injection Module (RSIM) is applied to RAMs produced by RAMM to obtain two Region Makeup Tensors, gamma and beta, which are subsequently added to the feature map of source image to transfer the makeup. As attention and makeup styles are calculated for each region, RamGAN can achieve better disentangled makeup transfer for different facial regions. When there are significant pose and expression variations between source and reference, RamGAN can also achieve better transfer results, due to the integration of spatial information and region-level correspondence. Experimental results are conducted on public datasets like MT, M-Wild and Makeup datasets, both visual and quantitative results and user study suggest that our approach achieves better transfer results than state-of-the-art methods like BeautyGAN, BeautyGlow, DMT, CPM and PSGAN.
Robust and accurate planar tracking over a whole video sequence is vitally important for many vision applications. The key to planar object tracking is to find object correspondences, modeled by homography, between the reference image and the tracked image. Existing methods tend to obtain wrong correspondences with changing appearance variations, camera-object relative motions and occlusions. To alleviate this problem, we present a unified convolutional neural network (CNN) model that jointly considers homography, visibility, and confidence. First, we introduce correlation blocks that explicitly account for the local appearance changes and camera-object relative motions as the base of our model. Second, we jointly learn the homography and visibility that links camera-object relative motions with occlusions. Third, we propose a confidence module that actively monitors the estimation quality from the pixel correlation distributions obtained in correlation blocks. All these modules are plugged into a Lucas-Kanade (LK) tracking pipeline to obtain both accurate and robust planar object tracking. Our approach outperforms the state-of-the-art methods on public POT and TMT datasets. Its superior performance is also verified on a real-world application, synthesizing high-quality in-video advertisements.
Understanding and forecasting future trajectories of agents are critical for behavior analysis, robot navigation, autonomous cars, and other related applications. Previous methods mostly treat trajectory prediction as time sequence generation. Different from them, this work studies agents’ trajectories in a ""vertical"" view, i.e., modeling and forecasting trajectories from the spectral domain. Different frequency bands in the trajectory spectrums could hierarchically reflect agents’ motion preferences at different scales. The low-frequency and high-frequency portions could represent their coarse motion trends and fine motion variations, respectively. Accordingly, we propose a hierarchical network V$^2$-Net, which contains two sub-networks, to hierarchically model and predict agents’ trajectories with trajectory spectrums. The coarse-level keypoints estimation sub-network first predicts the ""minimal"" spectrums of agents’ trajectories on several ""key"" frequency portions. Then the fine-level spectrum interpolation sub-network interpolates the spectrums to reconstruct the final predictions. Experimental results display the competitiveness and superiority of V$^2$-Net on both ETH-UCY benchmark and the Stanford Drone Dataset.
Knowing the 3D motions in a dynamic scene is essential to many vision applications. Recent progress is mainly focused on estimating the activity of some specific elements like humans. In this paper, we leverage a neural motion field for estimating the motion of all points in a multiview setting. Modeling the motion from a dynamic scene with multiview data is challenging due to the ambiguities in points of similar color and points with time-varying color. We propose to regularize the estimated motion to be predictable. If the motion from previous frames is known, then the motion in the near future should be predictable. Therefore, we introduce a predictability regularization by first conditioning the estimated motion on latent embeddings, then by adopting a predictor network to enforce predictability on the embeddings. The proposed framework PREF (Predictability REgularized Fields) achieves on par or better results than state-of-the-art neural motion field-based dynamic scene representation methods, while requiring no prior knowledge of the scene.