Naive DDIM inversion process usually suffers from a trajectory deviation issue, i.e., the latent trajectory during reconstruction deviates from the one during inversion. To alleviate this issue, previous methods either learn to mitigate the deviation or design cumbersome compensation strategy to reduce the mismatch error, exhibiting substantial time and computation cost. In this work, we present a nearly free-lunch method (named FreeInv) to address the issue more effectively and efficiently. In FreeInv, we randomly transform the latent representation and keep the transformation the same between the corresponding inversion and reconstruction time-step. It is motivated from a statistical perspective that an ensemble of DDIM inversion processes for multiple trajectories yields a smaller trajectory mismatch error on expectation. Moreover, through theoretical analysis and empirical study, we show that FreeInv performs an efficient ensemble of multiple trajectories. FreeInv can be freely integrated into existing inversion-based image and video editing techniques. Especially for inverting video sequences, it brings more significant fidelity and efficiency improvements. Comprehensive quantitative and qualitative evaluation on PIE benchmark and DAVIS dataset shows that FreeInv remarkably outperforms conventional DDIM inversion, and is competitive among previous state-of-the-art inversion methods, with superior computation efficiency.

Neural representations for 3D scenes have made substantial advancements recently, yet object removal remains a challenging yet practical issue, due to the absence of multi-view supervision over occluded areas. Diffusion Models (DMs), trained on extensive 2D images, show diverse and high-fidelity generative capabilities in the 2D domain. However, due to not being specifically trained on 3D data, their application to multi-view data often exacerbates inconsistency, hence impacting the overall quality of the 3D output. To address these issues, we introduce "In-N-Out", a novel approach that begins by inpainting a prior, i.e., the occluded area from a single view using DMs, followed by outstretching it to create multi-view inpaintings via latents alignments. Our analysis identifies that the variability in DMs' outputs mainly arises from initially sampled latents and intermediate latents predicted in the denoising process. We explicitly align of initial latents using a Neural Radiance Field (NeRF) to establish a consistent foundational structure in the inpainted area, complemented by an implicit alignment of intermediate latents through cross-view attention during the denoising phases, enhancing appearance consistency across views. To further enhance rendering results, we apply a patch-based hybrid loss to optimize NeRF. We demonstrate that our techniques effectively mitigate the challenges posed by inconsistencies in DMs and substantially improve the fidelity and coherence of inpainted 3D representations.

This paper studies how to flexibly integrate reconstructed 3D models into practical 3D modeling pipelines such as 3D scene creation and rendering. Due to the technical difficulty, one can only obtain rough 3D models (R3DMs) for most real objects using existing 3D reconstruction techniques. As a result, physically-based rendering (PBR) would render low-quality images or videos for scenes that are constructed by R3DMs. One promising solution would be representing real-world objects as Neural Fields such as NeRFs, which are able to generate photo-realistic renderings of an object under desired viewpoints. However, a drawback is that the synthesized views through Neural Fields Rendering (NFR) cannot reflect the simulated lighting details on R3DMs in PBR pipelines, especially when object interactions in the 3D scene creation cause local shadows. To solve this dilemma, we propose a lighting transfer network (LighTNet) to bridge NFR and PBR, such that they can benefit from each other. LighTNet reasons about a simplified image composition model, remedies the uneven surface issue caused by R3DMs, and is empowered by several perceptual-motivated constraints and a new Lab angle loss which enhances the contrast between lighting strength and colors. Comparisons demonstrate that LighTNet is superior in synthesizing impressive lighting, and is promising in pushing NFR further in practical 3D modeling workflows.

This paper studies implicit surface reconstruction leveraging differentiable ray casting. Previous works such as IDR [34] and NeuS [27] overlook the spatial context in 3D space when predicting and rendering the surface, thereby may fail to capture sharp local topologies such as small holes and structures. To mitigate the limitation, we propose a flexible neural implicit representation leveraging hierarchical voxel grids, namely Neural Deformable Anchor (NeuDA), for high-fidelity surface reconstruction. NeuDA maintains the hierarchical anchor grids where each vertex stores a 3D position (or anchor) instead of the direct embedding (or feature). We optimize the anchor grids such that different local geometry structures can be adaptively encoded. Besides, we dig into the frequency encoding strategies and introduce a simple hierarchical positional encoding method for the hierarchical anchor structure to flexibly exploit the properties of high-frequency and low-frequency geometry and appearance. Experiments on both the DTU [8] and BlendedMVS [32] datasets demonstrate that NeuDA can produce promising mesh surfaces.

The rendering scheme in neural radiance field (NeRF) is effective in rendering a pixel by casting a ray into the scene. However, NeRF yields blurred rendering results when the training images are captured at non-uniform scales, and produces aliasing artifacts if the test images are taken in distant views. To address this issue, Mip-NeRF proposes a multiscale representation as a conical frustum to encode scale information. Nevertheless, this approach is only suitable for offline rendering since it relies on integrated positional encoding (IPE) to query a multilayer perceptron (MLP). To overcome this limitation, we propose mip voxel grids (Mip-VoG), an explicit multiscale representation with a deferred architecture for real-time anti-aliasing rendering. Our approach includes a density Mip-VoG for scene geometry and a feature Mip-VoG with a small MLP for view-dependent color. Mip-VoG represents scene scale using the level of detail (LOD) derived from ray differentials and uses quadrilinear interpolation to map a queried 3D location to its features and density from two neighboring down-sampled voxel grids. To our knowledge, our approach is the first to offer multiscale training and real-time anti-aliasing rendering simultaneously. We conducted experiments on multiscale dataset, results show that our approach outperforms state-of-the-art real-time rendering baselines.

Unsupervised image-to-image (I21) translation aims to learn a domain mapping function that can preserve the semantics of the input images without paired data. However, because the underlying semantics distributions in the source and target domains are often mismatched, current distribution matching-based methods may distort the semantics when matching distributions, resulting in the inconsistency between the input and translated images, which is known as the semantics distortion problem. In this paper, we focus on the low-level I21 translation, where the structure of images is highly related to their semantics. To alleviate semantic distortions in such translation tasks without paired supervision, we propose a novel I21 translation constraint, called Structure Consistency Constraint (SCC), to promote the consistency of image structures by reducing the randomness of color transformation in the translation process. To facilitate estimation and maximization of SCC, we propose an approximate representation of mutual information called relative Squared-loss Mutual Information (rSMI) that enjoys efficient analytic solutions. Our SCC can be easily incorporated into most existing translation models. Quantitative and qualitative comparisons on a range of low-level I21 translation tasks show that translation models with SCC outperform the original models by a significant margin with little additional computational and memory costs.

Recent advances in implicit neural representations and differentiable rendering make it possible to simultaneously recover the geometry and materials of an object from multi-view RGB images captured under unknown static illumination. Despite the promising results achieved, indirect illumination is rarely modeled in previous methods, as it requires expensive recursive path tracing which makes the inverse rendering computationally intractable. In this paper, we propose a novel approach to efficiently recovering spatially-varying indirect illumination. The key insight is that indirect illumination can be conveniently derived from the neural radiance field learned from input images instead of being estimated jointly with direct illumination and materials. By properly modeling the indirect illumination and visibility of direct illumination, interreflection- and shadow-free albedo can be recovered. The experiments on both synthetic and real data demonstrate the superior performance of our approach compared to previous work and its capambility to synthesize realistic renderings under novel view-points and illumination. Our code and data are available at https://zju3dv.github.io/invrender/.

Neural Radiance Fields (NeRF) [22] have emerged as a potent paradigm for representing scenes and synthesizing photo-realistic images. A main limitation of conventional NeRFs is that they often fail to produce high-quality renderings under novel viewpoints that are significantly different from the training viewpoints. In this paper, instead of ex-ploiting few-shot image synthesis, we study the novel view extrapolation setting that (1) the training images can well describe an object, and (2) there is a notable discrepancy between the training and test viewpoints' distributions. We present RapNeRF (RAy Priors) as a solution. Our insight is that the inherent appearances of a 3D surface's arbitrary visible projections should be consistent. We thus propose a random ray casting policy that allows training unseen views using seen views. Furthermore, we show that a ray atlas pre-computed from the observed rays' viewing directions could further enhance the rendering quality for ex-trapolated views. A main limitation is that RapNeRF would remove the strong view-dependent effects because it lever-ages the multi-view consistency property.

Occlusion boundaries contain rich perceptual information about the underlying scene structure and provide important cues in many visual perception-related tasks such as object recognition, segmentation, motion estimation, scene understanding, and autonomous navigation. However, there is no formal definition of occlusion boundaries in the literature, and state-of-the-art occlusion boundary detection is still suboptimal. With this in mind, in this paper we propose a formal definition of occlusion boundaries for related studies. Further, based on a novel idea, we develop two concrete approaches with different characteristics to detect occlusion boundaries in video sequences via enhanced exploration of contextual information (e.g, local structural boundary patterns, observations from surrounding regions, and temporal context) with deep models and conditional random fields. Experimental evaluations of our methods on two challenging occlusion boundary benchmarks (CMU and VSB100) demonstrate that our detectors significantly outperform the current state-of-the-art. Finally, we empirically assess the roles of several important components of the proposed detectors to validate the rationale behind these approaches.

Adapting semantic segmentation models to new domains is an important but challenging problem. Recently enlightening progress has been made, but the performance of existing methods is unsatisfactory on real datasets where the new target domain comprises of heterogeneous sub-domains (e.g. diverse weather characteristics). We point out that carefully reasoning about the multiple modalities in the target domain can improve the robustness of adaptation models. To this end, we propose a condition-guided adaptation framework that is empowered by a special attentive progressive adversarial training (APAT) mechanism and a novel self-training policy. The APAT strategy progressively performs condition-specific alignment and attentive global feature matching. The new self-training scheme exploits the adversarial ambivalences of easy and hard adaptation regions and the correlations among target sub-domains effectively. We evaluate our method (DCAA) on various adaptation scenarios where the target images vary in weather conditions. The comparisons against baselines and the state-of-the-art approaches demonstrate the superiority of DCAA over the competitors.

We introduce 3D-FRONT (3D Furnished Rooms with layOuts and semaNTics), a new, large-scale, and comprehensive repository of synthetic indoor scenes highlighted by professionally designed layouts and a large number of rooms populated by high-quality textured 3D models with style compatibility. From layout semantics down to texture details of individual objects, our dataset is freely available to the academic community and beyond. Currently, 3D-FRONT contains 6,813 CAD houses, where 18,968 rooms diversely furnished by 3D objects, far surpassing all publicly available scene datasets. The 13,151 furniture objects all come with high-quality textures. While the floorplans and layout designs (i.e., furniture arrangements) are directly sourced from professional creations, the interior designs in terms of furniture styles, color, and textures have been carefully curated based on a recommender system we develop to attain consistent styles as expert designs. Furthermore, we release Trescope, a light-weight rendering tool, to support benchmark rendering of 2D images and annotations from 3D-FRONT. We demonstrate two applications, interior scene synthesis and texture synthesis, that are especially tailored to the strengths of our new dataset.

Depth completion aims to recover a dense depth map from the sparse depth data and the corresponding single RGB image. The observed pixels provide the significant guidance for the recovery of the unobserved pixels' depth. However, due to the sparsity of the depth data, the standard convolution operation, exploited by most of existing methods, is not effective to model the observed contexts with depth values. To address this issue, we propose to adopt the graph propagation to capture the observed spatial contexts. Specifically, we first construct multiple graphs at different scales from observed pixels. Since the graph structure varies from sample to sample, we then apply the attention mechanism on the propagation, which encourages the network to model the contextual information adaptively. Furthermore, considering the mutli-modality of input data, we exploit the graph propagation on the two modalities respectively to extract multi-modal representations. Finally, we introduce the symmetric gated fusion strategy to exploit the extracted multi-modal features effectively. The proposed strategy preserves the original information for one modality and also absorbs complementary information from the other through learning the adaptive gating weights. Our model, named Adaptive Context-Aware Multi-Modal Network (ACMNet), achieves the state-of-the-art performance on two benchmarks, i.e., KITTI and NYU-v2, and at the same time has fewer parameters than latest models. Our code is available at: https://github.com/sshan-zhao/ACMNet.

Image-based 3D shape retrieval (IBSR) aims to find the corresponding 3D shape of a given 2D image from a large 3D shape database. The common routine is to map 2D images and 3D shapes into an embedding space and define (or learn) a shape similarity measure. While metric learning with some adaptation techniques seems to be a natural solution to shape similarity learning, the performance is often unsatisfactory for fine-grained shape retrieval. In the paper, we identify the source of the poor performance and propose a practical solution to this problem. We find that the shape difference between a negative pair is entangled with the texture gap, making metric learning ineffective in pushing away negative pairs. To tackle this issue, we develop a geometry-focused multi-view metric learning framework empowered by texture synthesis. The synthesis of textures for 3D shape models creates hard triplets, which suppress the adverse effects of rich texture in 2D images, thereby push the network to focus more on discovering geometric characteristics. Our approach shows state-of-the-art performance on a recently released large-scale 3D-FUTURE [1] repository, as well as three widely studied benchmarks, including Pix3D [2], Stanford Cars [3], and Comp Cars [4]. Codes will be made publicly available at: https://github.com/3D-FRONT-FUTURE/IBSR-texture.

Domain generalization aims to learn from multiple source domains a predictive model that can generalize to unseen target domains. One essential problem in domain generalization is to learn discriminative domain-invariant features. To arrive at this, some methods introduce a domain discriminator through adversarial learning to match the feature distributions in multiple source domains. However, adversarial training can only guarantee that the learned features have invariant marginal distributions, while the invariance of conditional distributions is more important for prediction in new domains. To ensure the conditional invariance of learned features, we propose an entropy regularization term that measures the dependency between the learned features and the class labels. Combined with the typical task-related loss, e.g., cross-entropy loss for classification, and adversarial loss for domain discrimination, our overall objective is guaranteed to learn conditional-invariant features across all source domains and thus can learn classifiers with better generalization capabilities. We demonstrate the effectiveness of our method through comparison with state-of-the-art methods on both simulated and real-world datasets. Code is available at: https://github.com/sshan-zhao/DGviaER.

Unsupervised domain mapping aims to learn a function GXY to translate domain X to Y in the absence of paired examples. Finding the optimal GXY without paired data is an ill-posed problem, so appropriate constraints are required to obtain reasonable solutions. While some prominent constraints such as cycle consistency and distance preservation successfully constrain the solution space, they overlook the special properties of images that simple geometric transformations do not change the image's semantic structure. Based on this special property, we develop a geometry-consistent generative adversarial network (Gc-GAN), which enables one-sided unsupervised domain mapping. GcGAN takes the original image and its counterpart image transformed by a predefined geometric transformation as inputs and generates two images in the new domain coupled with the corresponding geometry-consistency constraint. The geometry-consistency constraint reduces the space of possible solutions while keep the correct solutions in the search space. Quantitative and qualitative comparisons with the baseline (GAN alone) and the state-of-the-art methods including CycleGAN [66] and DistanceGAN [5] demonstrate the effectiveness of our method.

Supervised depth estimation has achieved high accuracy due to the advanced deep network architectures. Since the groundtruth depth labels are hard to obtain, recent methods try to learn depth estimation networks in an unsupervised way by exploring unsupervised cues, which are effective but less reliable than true labels. An emerging way to resolve this dilemma is to transfer knowledge from synthetic images with ground truth depth via domain adaptation techniques. However, these approaches overlook specific geometric structure of the natural images in the target domain (i.e., real data), which is important for high-performing depth prediction. Motivated by the observation, we propose a geometry-aware symmetric domain adaptation framework (GASDA) to explore the labels in the synthetic data and epipolar geometry in the real data jointly. Moreover, by training two image style translators and depth estimators symmetrically in an end-to-end network, our model achieves better image style transfer and generates high-quality depth maps. The experimental results demonstrate the effectiveness of our proposed method and comparable performance against the state-of-the-art.

Monocular depth estimation, which plays a crucial role in understanding 3D scene geometry, is an ill-posed problem. Recent methods have gained significant improvement by exploring image-level information and hierarchical features from deep convolutional neural networks (DCNNs). These methods model depth estimation as a regression problem and train the regression networks by minimizing mean squared error, which suffers from slow convergence and unsatisfactory local solutions. Besides, existing depth estimation networks employ repeated spatial pooling operations, resulting in undesirable low-resolution feature maps. To obtain high-resolution depth maps, skip-connections or multilayer deconvolution networks are required, which complicates network training and consumes much more computations. To eliminate or at least largely reduce these problems, we introduce a spacing-increasing discretization (SID) strategy to discretize depth and recast depth network learning as an ordinal regression problem. By training the network using an ordinary regression loss, our method achieves much higher accuracy and faster convergence in synch. Furthermore, we adopt a multi-scale network structure which avoids unnecessary spatial pooling and captures multi-scale information in parallel. The proposed deep ordinal regression network (DORN) achieves state-of-the-art results on three challenging benchmarks, i.e., KITTI [16], Make3D [49], and NYU Depth v2 [41], and outperforms existing methods by a large margin.

The human visual system excels at detecting local blur of visual images, but the underlying mechanism is not well understood. Traditional views of blur such as reduction in energy at high frequencies and loss of phase coherence at localized features have fundamental limitations. For example, they cannot well discriminate flat regions from blurred ones. Here we propose that high-level semantic information is critical in successfully identifying local blur. Therefore, we resort to deep neural networks that are proficient at learning high-level features and propose the first end-to-end local blur mapping algorithm based on a fully convolutional network. By analyzing various architectures with different depths and design philosophies, we empirically show that high-level features of deeper layers play a more important role than low-level features of shallower layers in resolving challenging ambiguities for this task. We test the proposed method on a standard blur detection benchmark and demonstrate that it significantly advances the state-of-the-art (ODS F-score of 0.853). Furthermore, we explore the use of the generated blur maps in three applications, including blur region segmentation, blur degree estimation, and blur magnification.

Occlusion boundaries contain rich perceptual information about the underlying scene structure. They also provide important cues in many visual perception tasks such as scene understanding, object recognition, and segmentation. In this paper, we improve occlusion boundary detection via enhanced exploration of contextual information (e.g., local structural boundary patterns, observations from surrounding regions, and temporal context), and in doing so develop a novel approach based on convolutional neural networks (CNNs) and conditional random fields (CRFs). Experimental results demonstrate that our detector significantly outperforms the state-of-the-art (e.g., improving the F-measure from 0.62 to 0.71 on the commonly used CMU benchmark). Last but not least, we empirically assess the roles of several important components of the proposed detector, so as to validate the rationale behind this approach.

Facial expression recognition from thermal infrared images has attracted more and more attentions in recent years. However, the features adopted in current work are either temperature statistical parameters extracted from the facial regions of interest or several hand-crafted features that are commonly used in visible spectrum. Till now there is no image features specially defined for thermal infrared images. In this paper, we are the first to propose using the Deep Boltzmann Machine to learn thermal features for expression recognition from thermal long wavelength infrared images. First, the face are located and normalized from the thermal infrared images. Then, a Deep Boltzmann Machine model composed of two layers is proposed. The parameters of the Deep Boltzmann Machine model are further fine-tuned for facial expression recognition after pre-training of feature learning. Comparison experimental results on the NVIE database demonstrate that our approach outperforms other approaches using temperature statistic features or hand-crafted features borrowed from visible domain. The learned features from the forehead, mouth, and cheek are more reliable for discriminating disgust, fear, and happiness compared with other facial areas.