CVPR 2023 Papers

This paper addresses the challenge of efficiently reconstructing volumetric videos of dynamic humans from sparse multi-view videos. Some recent works represent a dynamic human as a canonical neural radiance field (NeRF) and a motion field, which are learned from input videos through differentiable rendering. But the per-scene optimization generally requires hours. Other generalizable NeRF models leverage learned prior from datasets to reduce the optimization time by only finetuning on new scenes at the cost of visual fidelity. In this paper, we propose a novel method for learning neural volumetric representations of dynamic humans in minutes with competitive visual quality. Specifically, we define a novel part-based voxelized human representation to better distribute the representational power of the network to different human parts. Furthermore, we propose a novel 2D motion parameterization scheme to increase the convergence rate of deformation field learning. Experiments demonstrate that our model can be learned 100 times faster than previous per-scene optimization methods while being competitive in the rendering quality. Training our model on a 512x512 video with 100 frames typically takes about 5 minutes on a single RTX 3090 GPU. The code is available on our project page: https://zju3dv.github.io/instant_nvr

Generative models show good potential for recovering 3D faces beyond limited shape assumptions. While plausible details and resolutions are achieved, these models easily fail under extreme conditions of pose, shadow or appearance, due to the entangled fitting or lack of multi-view priors. To address this problem, this paper presents a novel Neural Proto-face Field (NPF) for unsupervised robust 3D face modeling. Instead of using constrained images as Neural Radiance Field (NeRF), NPF disentangles the common/specific facial cues, i.e., ID, expression and scene-specific details from in-the-wild photo collections. Specifically, NPF learns a face prototype to aggregate 3D-consistent identity via uncertainty modeling, extracting multi-image priors from a photo collection. NPF then learns to deform the prototype with the appropriate facial expressions, constrained by a loss of expression consistency and personal idiosyncrasies. Finally, NPF is optimized to fit a target image in the collection, recovering specific details of appearance and geometry. In this way, the generative model benefits from multi-image priors and meaningful facial structures. Extensive experiments on benchmarks show that NPF recovers superior or competitive facial shapes and textures, compared to state-of-the-art methods.

Neural radiance fields (NeRFs) enable novel view synthesis with unprecedented visual quality. However, to render photorealistic images, NeRFs require hundreds of deep multilayer perceptron (MLP) evaluations -- for each pixel. This is prohibitively expensive and makes real-time rendering infeasible, even on powerful modern GPUs. In this paper, we propose a novel approach to distill and bake NeRFs into highly efficient mesh-based neural representations that are fully compatible with the massively parallel graphics rendering pipeline. We represent scenes as neural radiance features encoded on a two-layer duplex mesh, which effectively overcomes the inherent inaccuracies in 3D surface reconstruction by learning the aggregated radiance information from a reliable interval of ray-surface intersections. To exploit local geometric relationships of nearby pixels, we leverage screen-space convolutions instead of the MLPs used in NeRFs to achieve high-quality appearance. Finally, the performance of the whole framework is further boosted by a novel multi-view distillation optimization strategy. We demonstrate the effectiveness and superiority of our approach via extensive experiments on a range of standard datasets.

The real-world data tends to be heavily imbalanced and severely skew the data-driven deep neural networks, which makes Long-Tailed Recognition (LTR) a massive challenging task. Existing LTR methods seldom train Vision Transformers (ViTs) with Long-Tailed (LT) data, while the off-the-shelf pretrain weight of ViTs always leads to unfair comparisons. In this paper, we systematically investigate the ViTs' performance in LTR and propose LiVT to train ViTs from scratch only with LT data. With the observation that ViTs suffer more severe LTR problems, we conduct Masked Generative Pretraining (MGP) to learn generalized features. With ample and solid evidence, we show that MGP is more robust than supervised manners. Although Binary Cross Entropy (BCE) loss performs well with ViTs, it struggles on the LTR tasks. We further propose the balanced BCE to ameliorate it with strong theoretical groundings. Specially, we derive the unbiased extension of Sigmoid and compensate extra logit margins for deploying it. Our Bal-BCE contributes to the quick convergence of ViTs in just a few epochs. Extensive experiments demonstrate that with MGP and Bal-BCE, LiVT successfully trains ViTs well without any additional data and outperforms comparable state-of-the-art methods significantly, e.g., our ViT-B achieves 81.0% Top-1 accuracy in iNaturalist 2018 without bells and whistles. Code is available at https://github.com/XuZhengzhuo/LiVT.

Prompt learning is an efficient approach to adapt transformers by inserting learnable set of parameters into the input and intermediate representations of a pre-trained model. In this work, we present Expressive Prompts with Residuals (EXPRES) which modifies the prompt learning paradigm specifically for effective adaptation of vision transformers (ViT). Out method constructs downstream representations via learnable "output" tokens, that are akin to the learned class tokens of the ViT. Further for better steering of the downstream representation processed by the frozen transformer, we introduce residual learnable tokens that are added to the output of various computations. We apply EXPRES for image classification, few shot learning, and semantic segmentation, and show our method is capable of achieving state of the art prompt tuning on 3/3 categories of the VTAB benchmark. In addition to strong performance, we observe that our approach is an order of magnitude more prompt efficient than existing visual prompting baselines. We analytically show the computational benefits of our approach over weight space adaptation techniques like finetuning. Lastly we systematically corroborate the architectural design of our method via a series of ablation experiments.

A key challenge for novel view synthesis of monocular portrait images is 3D consistency under continuous pose variations. Most existing methods rely on 2D generative models which often leads to obvious 3D inconsistency artifacts. We present a 3D-consistent novel view synthesis approach for monocular portrait images based on a recent proposed 3D-aware GAN, namely Generative Radiance Manifolds (GRAM), which has shown strong 3D consistency at multiview image generation of virtual subjects via the radiance manifolds representation. However, simply learning an encoder to map a real image into the latent space of GRAM can only reconstruct coarse radiance manifolds without faithful fine details, while improving the reconstruction fidelity via instance-specific optimization is time-consuming. We introduce a novel detail manifolds reconstructor to learn 3D-consistent fine details on the radiance manifolds from monocular images, and combine them with the coarse radiance manifolds for high-fidelity reconstruction. The 3D priors derived from the coarse radiance manifolds are used to regulate the learned details to ensure reasonable synthesized results at novel views. Trained on in-the-wild 2D images, our method achieves high-fidelity and 3D-consistent portrait synthesis largely outperforming the prior art. Project page: https://yudeng.github.io/GRAMInverter/

LiDAR provides accurate geometric measurements of the 3D world. Unfortunately, dense LiDARs are very expensive and the point clouds captured by low-beam LiDAR are often sparse. To address these issues, we present UltraLiDAR, a data-driven framework for scene-level LiDAR completion, LiDAR generation, and LiDAR manipulation. The crux of UltraLiDAR is a compact, discrete representation that encodes the point cloud's geometric structure, is robust to noise, and is easy to manipulate. We show that by aligning the representation of a sparse point cloud to that of a dense point cloud, we can densify the sparse point clouds as if they were captured by a real high-density LiDAR, drastically reducing the cost. Furthermore, by learning a prior over the discrete codebook, we can generate diverse, realistic LiDAR point clouds for self-driving. We evaluate the effectiveness of UltraLiDAR on sparse-to-dense LiDAR completion and LiDAR generation. Experiments show that densifying real-world point clouds with our approach can significantly improve the performance of downstream perception systems. Compared to prior art on LiDAR generation, our approach generates much more realistic point clouds. According to A/B test, over 98.5% of the time human participants prefer our results over those of previous methods. Please refer to project page https://waabi.ai/research/ultralidar/ for more information.

This paper shows that it is possible to learn models for monocular 3D reconstruction of articulated objects (e.g. horses, cows, sheep), using as few as 50-150 images labeled with 2D keypoints. Our proposed approach involves training category-specific keypoint estimators, generating 2D keypoint pseudo-labels on unlabeled web images, and using both the labeled and self-labeled sets to train 3D reconstruction models. It is based on two key insights: (1) 2D keypoint estimation networks trained on as few as 50-150 images of a given object category generalize well and generate reliable pseudo-labels; (2) a data selection mechanism can automatically create a "curated" subset of the unlabeled web images that can be used for training -- we evaluate four data selection methods. Coupling these two insights enables us to train models that effectively utilize web images, resulting in improved 3D reconstruction performance for several articulated object categories beyond the fully-supervised baseline. Our approach can quickly bootstrap a model and requires only a few images labeled with 2D keypoints. This requirement can be easily satisfied for any new object category. To showcase the practicality of our approach for predicting the 3D shape of arbitrary object categories, we annotate 2D keypoints on 250 giraffe and bear images from COCO in just 2.5 hours per category.

Pre-training by numerous image data has become de-facto for robust 2D representations. In contrast, due to the expensive data processing, a paucity of 3D datasets severely hinders the learning for high-quality 3D features. In this paper, we propose an alternative to obtain superior 3D representations from 2D pre-trained models via Image-to-Point Masked Autoencoders, named as I2P-MAE. By self-supervised pre-training, we leverage the well learned 2D knowledge to guide 3D masked autoencoding, which reconstructs the masked point tokens with an encoder-decoder architecture. Specifically, we first utilize off-the-shelf 2D models to extract the multi-view visual features of the input point cloud, and then conduct two types of image-to-point learning schemes. For one, we introduce a 2D-guided masking strategy that maintains semantically important point tokens to be visible. Compared to random masking, the network can better concentrate on significant 3D structures with key spatial cues. For another, we enforce these visible tokens to reconstruct multi-view 2D features after the decoder. This enables the network to effectively inherit high-level 2D semantics for discriminative 3D modeling. Aided by our image-to-point pre-training, the frozen I2P-MAE, without any fine-tuning, achieves 93.4% accuracy for linear SVM on ModelNet40, competitive to existing fully trained methods. By further fine-tuning on on ScanObjectNN's hardest split, I2P-MAE attains the state-of-the-art 90.11% accuracy, +3.68% to the second-best, demonstrating superior transferable capacity. Code is available at https://github.com/ZrrSkywalker/I2P-MAE.

Point cloud sampling is crucial for efficient large-scale point cloud analysis, where learning-to-sample methods have recently received increasing attention from the community for jointly training with downstream tasks. However, the above-mentioned task-specific sampling methods usually fail to explore the geometries of objects in an explicit manner. In this paper, we introduce a new skeleton-aware learning-to-sample method by learning object skeletons as the prior knowledge to preserve the object geometry and topology information during sampling. Specifically, without labor-intensive annotations per object category, we first learn category-agnostic object skeletons via the medial axis transform definition in an unsupervised manner. With object skeleton, we then evaluate the histogram of the local feature size as the prior knowledge to formulate skeleton-aware sampling from a probabilistic perspective. Additionally, the proposed skeleton-aware sampling pipeline with the task network is thus end-to-end trainable by exploring the reparameterization trick. Extensive experiments on three popular downstream tasks, point cloud classification, retrieval, and reconstruction, demonstrate the effectiveness of the proposed method for efficient point cloud analysis.

Text-guided image generation has progressed rapidly in recent years, inspiring major breakthroughs in text-guided shape generation. Recently, it has been shown that using score distillation, one can successfully text-guide a NeRF model to generate a 3D object. We adapt the score distillation to the publicly available, and computationally efficient, Latent Diffusion Models, which apply the entire diffusion process in a compact latent space of a pretrained autoencoder. As NeRFs operate in image space, a naive solution for guiding them with latent score distillation would require encoding to the latent space at each guidance step. Instead, we propose to bring the NeRF to the latent space, resulting in a Latent-NeRF. Analyzing our Latent-NeRF, we show that while Text-to-3D models can generate impressive results, they are inherently unconstrained and may lack the ability to guide or enforce a specific 3D structure. To assist and direct the 3D generation, we propose to guide our Latent-NeRF using a Sketch-Shape: an abstract geometry that defines the coarse structure of the desired object. Then, we present means to integrate such a constraint directly into a Latent-NeRF. This unique combination of text and shape guidance allows for increased control over the generation process. We also show that latent score distillation can be successfully applied directly on 3D meshes. This allows for generating high-quality textures on a given geometry. Our experiments validate the power of our different forms of guidance and the efficiency of using latent rendering.

Deep Implicit Function (DIF) has gained much popularity as an efficient 3D shape representation. To capture geometry details, current mainstream methods divide 3D shapes into local regions and then learn each one with a local latent code via a decoder, where the decoder shares the geometric similarities among different local regions. Although such local methods can capture more local details, a large diversity of different local regions increases the difficulty of learning an implicit function when treating all regions equally using only a single decoder. In addition, these local regions often exhibit imbalanced distributions, where certain regions have significantly fewer observations. This leads that fine geometry details could not be preserved well. To solve this problem, we propose a novel Local Pattern-specific Implicit Function, named LP-DIF, for representing a shape with some clusters of local regions and multiple decoders, where each decoder only focuses on one cluster of local regions which share a certain pattern. Specifically, we first extract local codes for all regions, and then cluster them into multiple groups in the latent space, where similar regions sharing a common pattern fall into one group. After that, we train multiple decoders for mining local patterns of different groups, which simplifies learning of fine geometric details by reducing the diversity of local regions seen by each decoder. To further alleviate the data-imbalance problem, we introduce a region re-weighting module to each pattern-specific decoder by kernel density estimator, which dynamically re-weights the regions during learning. Our LP-DIF can restore more geometry details, and thus improve the quality of 3D reconstruction. Experiments demonstrate that our method can achieve the state-of-the-art performance over previous methods. Code is available at https://github.com/gtyxyz/lpdif.

Soft prompt learning has recently emerged as one of the methods of choice for adapting V&L models to a downstream task using a few training examples. However, current methods significantly overfit the training data, suffering from large accuracy degradation when tested on unseen classes from the same domain. To this end, in this paper, we make the following 4 contributions: (1) To alleviate base class overfitting, we propose a novel Language-Aware Soft Prompting (LASP) learning method by means of a text-to-text cross-entropy loss that maximizes the probability of the learned prompts to be correctly classified with respect to pre-defined hand-crafted textual prompts. (2) To increase the representation capacity of the prompts, we propose grouped LASP where each group of prompts is optimized with respect to a separate subset of textual prompts. (3) We identify a visual-language misalignment introduced by prompt learning and LASP, and more importantly, propose a re-calibration mechanism to address it. (4) We show that LASP is inherently amenable to including, during training, virtual classes, i.e. class names for which no visual samples are available, further increasing the robustness of the learned prompts. Through evaluations on 11 datasets, we show that our approach (a) significantly outperforms all prior works on soft prompting, and (b) matches and surpasses, for the first time, the accuracy on novel classes obtained by hand-crafted prompts and CLIP for 8 out of 11 test datasets. Code will be made available.

Although vision transformers (ViTs) have shown promising results in various computer vision tasks recently, their high computational cost limits their practical applications. Previous approaches that prune redundant tokens have demonstrated a good trade-off between performance and computation costs. Nevertheless, errors caused by pruning strategies can lead to significant information loss. Our quantitative experiments reveal that the impact of pruned tokens on performance should be noticeable. To address this issue, we propose a novel joint Token Pruning & Squeezing module (TPS) for compressing vision transformers with higher efficiency. Firstly, TPS adopts pruning to get the reserved and pruned subsets. Secondly, TPS squeezes the information of pruned tokens into partial reserved tokens via the unidirectional nearest-neighbor matching and similarity-oriented fusing steps. Compared to state-of-the-art methods, our approach outperforms them under all token pruning intensities. Especially while shrinking DeiT-tiny&small computational budgets to 35%, it improves the accuracy by 1%-6% compared with baselines on ImageNet classification. The proposed method can accelerate the throughput of DeiT-small beyond DeiT-tiny, while its accuracy surpasses DeiT-tiny by 4.78%. Experiments on various transformers demonstrate the effectiveness of our method, while analysis experiments prove our higher robustness to the errors of the token pruning policy. Code is available at https://github.com/megvii-research/TPS-CVPR2023.

We propose a method that trains a neural radiance field (NeRF) to encode not only the appearance of the scene but also semantic correlations between scene points, regions, or entities -- aiming to capture their mutual co-variation patterns. In contrast to the traditional first-order photometric reconstruction objective, our method explicitly regularizes the learning dynamics to align the Jacobians of highly-correlated entities, which proves to maximize the mutual information between them under random scene perturbations. By paying attention to this second-order information, we can shape a NeRF to express semantically meaningful synergies when the network weights are changed by a delta along the gradient of a single entity, region, or even a point. To demonstrate the merit of this mutual information modeling, we leverage the coordinated behavior of scene entities that emerges from our shaping to perform label propagation for semantic and instance segmentation. Our experiments show that a JacobiNeRF is more efficient in propagating annotations among 2D pixels and 3D points compared to NeRFs without mutual information shaping, especially in extremely sparse label regimes -- thus reducing annotation burden. The same machinery can further be used for entity selection or scene modifications. Our code is available at https://github.com/xxm19/jacobinerf.

This paper presents JAWS, an optimzation-driven approach that achieves the robust transfer of visual cinematic features from a reference in-the-wild video clip to a newly generated clip. To this end, we rely on an implicit-neural-representation (INR) in a way to compute a clip that shares the same cinematic features as the reference clip. We propose a general formulation of a camera optimization problem in an INR that computes extrinsic and intrinsic camera parameters as well as timing. By leveraging the differentiability of neural representations, we can back-propagate our designed cinematic losses measured on proxy estimators through a NeRF network to the proposed cinematic parameters directly. We also introduce specific enhancements such as guidance maps to improve the overall quality and efficiency. Results display the capacity of our system to replicate well known camera sequences from movies, adapting the framing, camera parameters and timing of the generated video clip to maximize the similarity with the reference clip.

The quality of point clouds is often limited by noise introduced during their capture process. Consequently, a fundamental 3D vision task is the removal of noise, known as point cloud filtering or denoising. State-of-the-art learning based methods focus on training neural networks to infer filtered displacements and directly shift noisy points onto the underlying clean surfaces. In high noise conditions, they iterate the filtering process. However, this iterative filtering is only done at test time and is less effective at ensuring points converge quickly onto the clean surfaces. We propose IterativePFN (iterative point cloud filtering network), which consists of multiple IterationModules that model the true iterative filtering process internally, within a single network. We train our IterativePFN network using a novel loss function that utilizes an adaptive ground truth target at each iteration to capture the relationship between intermediate filtering results during training. This ensures that the filtered results converge faster to the clean surfaces. Our method is able to obtain better performance compared to state-of-the-art methods. The source code can be found at: https://github.com/ddsediri/IterativePFN.

We present Iterative Vision-and-Language Navigation (IVLN), a paradigm for evaluating language-guided agents navigating in a persistent environment over time. Existing Vision-and-Language Navigation (VLN) benchmarks erase the agent's memory at the beginning of every episode, testing the ability to perform cold-start navigation with no prior information. However, deployed robots occupy the same environment for long periods of time. The IVLN paradigm addresses this disparity by training and evaluating VLN agents that maintain memory across tours of scenes that consist of up to 100 ordered instruction-following Room-to-Room (R2R) episodes, each defined by an individual language instruction and a target path. We present discrete and continuous Iterative Room-to-Room (IR2R) benchmarks comprising about 400 tours each in 80 indoor scenes. We find that extending the implicit memory of high-performing transformer VLN agents is not sufficient for IVLN, but agents that build maps can benefit from environment persistence, motivating a renewed focus on map-building agents in VLN.

In this work, we propose an inverse rendering model that estimates 3D shape, spatially-varying reflectance, homogeneous subsurface scattering parameters, and an environment illumination jointly from only a pair of captured images of a translucent object. In order to solve the ambiguity problem of inverse rendering, we use a physically-based renderer and a neural renderer for scene reconstruction and material editing. Because two renderers are differentiable, we can compute a reconstruction loss to assist parameter estimation. To enhance the supervision of the proposed neural renderer, we also propose an augmented loss. In addition, we use a flash and no-flash image pair as the input. To supervise the training, we constructed a large-scale synthetic dataset of translucent objects, which consists of 117K scenes. Qualitative and quantitative results on both synthetic and real-world datasets demonstrated the effectiveness of the proposed model.