CVPR 2024 Papers

We investigate the impact of deep generative models on potential social biases in upcoming computer vision models. As the internet witnesses an increasing influx of AI-generated images concerns arise regarding inherent biases that may accompany them potentially leading to the dissemination of harmful content. This paper explores whether a detrimental feedback loop resulting in bias amplification would occur if generated images were used as the training data for future models. We conduct simulations by progressively substituting original images in COCO and CC3M datasets with images generated through Stable Diffusion. The modified datasets are used to train OpenCLIP and image captioning models which we evaluate in terms of quality and bias. Contrary to expectations our findings indicate that introducing generated images during training does not uniformly amplify bias. Instead instances of bias mitigation across specific tasks are observed. We further explore the factors that may influence these phenomena such as artifacts in image generation (e.g. blurry faces) or pre-existing biases in the original datasets.

Creating multi-view wire art (MVWA) a static 3D sculpture with diverse interpretations from different viewpoints is a complex task even for skilled artists. In response we present DreamWire an AI system enabling everyone to craft MVWA easily. Users express their vision through text prompts or scribbles freeing them from intricate 3D wire organisation. Our approach synergises 3D Bezier curves Prim's algorithm and knowledge distillation from diffusion models or their variants (e.g. ControlNet). This blend enables the system to represent 3D wire art ensuring spatial continuity and overcoming data scarcity. Extensive evaluation and analysis are conducted to shed insight on the inner workings of the proposed system including the trade-off between connectivity and visual aesthetics.

Generative Zero-shot learning (ZSL) learns a generator to synthesize visual samples for unseen classes which is an effective way to advance ZSL. However existing generative methods rely on the conditions of Gaussian noise and the predefined semantic prototype which limit the generator only optimized on specific seen classes rather than characterizing each visual instance resulting in poor generalizations (e.g. overfitting to seen classes). To address this issue we propose a novel Visual-Augmented Dynamic Semantic prototype method (termed VADS) to boost the generator to learn accurate semantic-visual mapping by fully exploiting the visual-augmented knowledge into semantic conditions. In detail VADS consists of two modules: (1) Visual-aware Domain Knowledge Learning module (VDKL) learns the local bias and global prior of the visual features (referred to as domain visual knowledge) which replace pure Gaussian noise to provide richer prior noise information; (2) Vision-Oriented Semantic Updation module (VOSU) updates the semantic prototype according to the visual representations of the samples. Ultimately we concatenate their output as a dynamic semantic prototype which serves as the condition of the generator. Extensive experiments demonstrate that our VADS achieves superior CZSL and GZSL performances on three prominent datasets and outperforms other state-of-the-art methods with averaging increases by 6.4% 5.9% and 4.2% on SUN CUB and AWA2 respectively.

Recently the advancement of self-supervised learning techniques like masked autoencoders (MAE) has greatly influenced visual representation learning for images and videos. Nevertheless it is worth noting that the predominant approaches in existing masked image / video modeling rely excessively on resource-intensive vision transformers (ViTs) as the feature encoder. In this paper we propose a new approach termed as VideoMAC which combines video masked autoencoders with resource-friendly ConvNets. Specifically VideoMAC employs symmetric masking on randomly sampled pairs of video frames. To prevent the issue of mask pattern dissipation we utilize ConvNets which are implemented with sparse convolutional operators as encoders. Simultaneously we present a simple yet effective masked video modeling (MVM) approach a dual encoder architecture comprising an online encoder and an exponential moving average target encoder aimed to facilitate inter-frame reconstruction consistency in videos. Additionally we demonstrate that VideoMAC empowering classical (ResNet) / modern (ConvNeXt) convolutional encoders to harness the benefits of MVM outperforms ViT-based approaches on downstream tasks including video object segmentation (+5.2% / 6.4% \mathcal J &\mathcal F ) body part propagation (+6.3% / 3.1% mIoU) and human pose tracking (+10.2% / 11.1% PCK@0.1).

Video generation has witnessed significant advancements yet evaluating these models remains a challenge. A comprehensive evaluation benchmark for video generation is indispensable for two reasons: 1) Existing metrics do not fully align with human perceptions; 2) An ideal evaluation system should provide insights to inform future developments of video generation. To this end we present VBench a comprehensive benchmark suite that dissects "video generation quality" into specific hierarchical and disentangled dimensions each with tailored prompts and evaluation methods. VBench has three appealing properties: 1) Comprehensive Dimensions: VBench comprises 16 dimensions in video generation (e.g. subject identity inconsistency motion smoothness temporal flickering and spatial relationship etc). The evaluation metrics with fine-grained levels reveal individual models' strengths and weaknesses. 2) Human Alignment: We also provide a dataset of human preference annotations to validate our benchmarks' alignment with human perception for each evaluation dimension respectively. 3) Valuable Insights: We look into current models' ability across various evaluation dimensions and various content types. We also investigate the gaps between video and image generation models. We will open-source VBench including all prompts evaluation methods generated videos and human preference annotations and also include more video generation models in VBench to drive forward the field of video generation.

The booming use of text-to-image generative models has raised concerns about their high risk of producing copyright-infringing content. While probabilistic copyright protection methods provide a probabilistic guarantee against such infringement in this paper we introduce Virtually Assured Amplification Attack (VA3) a novel online attack framework that exposes the vulnerabilities of these protection mechanisms. The proposed framework significantly amplifies the probability of generating infringing content on the sustained interactions with generative models and a non-trivial lower-bound on the success probability of each engagement. Our theoretical and experimental results demonstrate the effectiveness of our approach under various scenarios.These findings highlight the potential risk of implementing probabilistic copyright protection in practical applications of text-to-image generative models. Code is available at https://github.com/South7X/VA3.

Unsupervised learning of keypoints and landmarks has seen significant progress with the help of modern neural network architectures but performance is yet to match the supervised counterpart making their practicability questionable. We leverage the emergent knowledge within text-to-image diffusion models towards more robust unsupervised keypoints. Our core idea is to find text embeddings that would cause the generative model to consistently attend to compact regions in images (i.e. keypoints). To do so we simply optimize the text embedding such that the cross-attention maps within the denoising network are localized as Gaussians with small standard deviations. We validate our performance on multiple datasets: the CelebA CUB-200-2011 Tai-Chi-HD DeepFashion and Human3.6m datasets. We achieve significantly improved accuracy sometimes even outperforming supervised ones particularly for data that is non-aligned and less curated. Our code is publicly available at https://stablekeypoints.github.io/.

Generalizable neural implicit surface reconstruction aims to obtain an accurate underlying geometry given a limited number of multi-view images from unseen scenes. However existing methods select only informative and relevant views using predefined scores for training and testing phases. This constraint renders the model impractical in real-world scenarios where the availability of favorable combinations cannot always be ensured. We introduce and validate a view-combination score to indicate the effectiveness of the input view combination. We observe that previous methods output degenerate solutions under arbitrary and unfavorable sets. Building upon this finding we propose UFORecon a robust view-combination generalizable surface reconstruction framework. To achieve this we apply cross-view matching transformers to model interactions between source images and build correlation frustums to capture global correlations. Additionally we explicitly encode pairwise feature similarities as view-consistent priors. Our proposed framework significantly outperforms previous methods in terms of view-combination generalizability and also in the conventional generalizable protocol trained with favorable view-combinations. The code is available at https://github.com/Youngju-Na/UFORecon.

Super-resolution (SR) is an ill-posed inverse problem where the size of the set of feasible solutions that are consistent with a given low-resolution image is very large. Many algorithms have been proposed to find a "good" solution among the feasible solutions that strike a balance between fidelity and perceptual quality. Unfortunately all known methods generate artifacts and hallucinations while trying to reconstruct high-frequency (HF) image details. A fundamental question is: Can a model learn to distinguish genuine image details from artifacts? Although some recent works focused on the differentiation of details and artifacts this is a very challenging problem and a satisfactory solution is yet to be found. This paper shows that the characterization of genuine HF details versus artifacts can be better learned by training GAN-based SR models using wavelet-domain loss functions compared to RGB-domain or Fourier-space losses. Although wavelet-domain losses have been used in the literature before they have not been used in the context of the SR task. More specifically we train the discriminator only on the HF wavelet sub-bands instead of on RGB images and the generator is trained by a fidelity loss over wavelet subbands to make it sensitive to the scale and orientation of structures. Extensive experimental results demonstrate that our model achieves better perception-distortion trade-off according to multiple objective measures and visual evaluations.

We present HAAR a new strand-based generative model for 3D human hairstyles. Specifically based on textual inputs HAAR produces 3D hairstyles that could be used as production-level assets in modern computer graphics engines. Current AI-based generative models take advantage of powerful 2D priors to reconstruct 3D content in the form of point clouds meshes or volumetric functions. However by using the 2D priors they are intrinsically limited to only recovering the visual parts. Highly occluded hair structures can not be reconstructed with those methods and they only model the "outer shell" which is not ready to be used in physics-based rendering or simulation pipelines. In contrast we propose a first text-guided generative method that uses 3D hair strands as an underlying representation. Leveraging 2D visual question-answering (VQA) systems we automatically annotate synthetic hair models that are generated from a small set of artist-created hairstyles. This allows us to train a latent diffusion model that operates in a common hairstyle UV space. In qualitative and quantitative studies we demonstrate the capabilities of the proposed model and compare it to existing hairstyle generation approaches. For results please refer to our project page https://haar.is.tue.mpg.de/.

We present a scene representation that brings vision and touch into a shared 3D space which we call a tactile-augmented radiance field. This representation capitalizes on two key insights: (i) ubiquitous vision-based touch sensors are built on perspective cameras and (ii) visually and structurally similar regions of a scene share the same tactile features. We use these insights to train a conditional diffusion model that provided with an RGB image and a depth map rendered from a neural radiance field generates its corresponding tactile "image". To train this diffusion model we collect the largest collection of spatially-aligned visual and tactile data. Through qualitative and quantitative experiments we demonstrate the accuracy of our cross-modal generative model and the utility of collected and rendered visual-tactile pairs across a range of downstream tasks. Project page: https://dou-yiming.github.io/TaRF

A neural signed distance function (SDF) is a convenient shape representation for many tasks such as surface reconstruction editing and generation. However neural SDFs are difficult to fit to raw point clouds such as those sampled from the surface of a shape by a scanner. A major issue occurs when the shape's geometry is very different from the structural biases implicit in the network's initialization. In this case we observe that the standard loss formulation does not guide the network towards the correct SDF values. We circumvent this problem by introducing guiding points and use them to steer the optimization towards the true shape via small incremental changes for which the loss formulation has a good descent direction. We show that this point-guided homotopy-based optimization scheme facilitates a deformation from an easy problem to the difficult reconstruction problem. We also propose a metric to quantify the difference in surface geometry between a target shape and an initial surface which helps indicate whether the standard loss formulation is guiding towards the target shape. Our method outperforms previous state-of-the-art approaches with large improvements on shapes identified by this metric as particularly challenging.

Editing 3D shapes through natural language instructions is a challenging task that requires the comprehension of both language semantics and fine-grained geometric details. To bridge this gap we introduce ShapeWalk a carefully designed synthetic dataset designed to advance the field of language-guided shape editing. The dataset consists of 158K unique shapes connected through 26K edit chains with an average length of 14 chained shapes. Each consecutive pair of shapes is associated with precise language instructions describing the applied edits. We synthesize edit chains by reconstructing and interpolating shapes sampled from a realistic CAD-designed 3D dataset in the parameter space of the GeoCode shape program. We leverage rule-based methods and language models to generate accurate and realistic natural language prompts corresponding to each edit. To illustrate the practicality of our contribution we train neural editor modules in the latent space of shape autoencoders and demonstrate the ability of our dataset to enable a variety of language-guided shape edits. Finally we introduce multi-step editing metrics to benchmark the capacity of our models to perform recursive shape edits. We hope that our work will enable further study of compositional language-guided shape editing and finds application in 3D CAD design and interactive modeling.

Generative models can produce impressively realistic images. This paper demonstrates that generated images have geometric features different from those of real images. We build a set of collections of generated images prequalified to fool simple signal-based classifiers into believing they are real. We then show that prequalified generated images can be identified reliably by classifiers that only look at geometric properties. We use three such classifiers. All three classifiers are denied access to image pixels and look only at derived geometric features. The first classifier looks at the perspective field of the image the second looks at lines detected in the image and the third looks at relations between detected objects and shadows. Our procedure detects generated images more reliably than SOTA local signal based detectors for images from a number of distinct generators. Saliency maps suggest that the classifiers can identify geometric problems reliably. We conclude that current generators cannot reliably reproduce geometric properties of real images.

In the field of class incremental learning (CIL) generative replay has become increasingly prominent as a method to mitigate the catastrophic forgetting alongside the continuous improvements in generative models. However its application in class incremental object detection (CIOD) has been significantly limited primarily due to the complexities of scenes involving multiple labels. In this paper we propose a novel approach called stable diffusion deep generative replay (SDDGR) for CIOD. Our method utilizes a diffusion-based generative model with pre-trained text-to-image diffusion networks to generate realistic and diverse synthetic images. SDDGR incorporates an iterative refinement strategy to produce high-quality images encompassing old classes. Additionally we adopt an L2 knowledge distillation technique to improve the retention of prior knowledge in synthetic images. Furthermore our approach includes pseudo-labeling for old objects within new task images preventing misclassification as background elements. Extensive experiments on the COCO 2017 dataset demonstrate that SDDGR significantly outperforms existing algorithms achieving a new state-of-the-art in various CIOD scenarios.

We present SCULPT a novel 3D generative model for clothed and textured 3D meshes of humans. Specifically we devise a deep neural network that learns to represent the geometry and appearance distribution of clothed human bodies. Training such a model is challenging as datasets of textured 3D meshes for humans are limited in size and accessibility. Our key observation is that there exist medium-sized 3D scan datasets like CAPE as well as large-scale 2D image datasets of clothed humans and multiple appearances can be mapped to a single geometry. To effectively learn from the two data modalities we propose an unpaired learning procedure for pose-dependent clothed and textured human meshes. Specifically we learn a pose-dependent geometry space from 3D scan data. We represent this as per vertex displacements w.r.t. the SMPL model. Next we train a geometry conditioned texture generator in an unsupervised way using the 2D image data. We use intermediate activations of the learned geometry model to condition our texture generator. To alleviate entanglement between pose and clothing type and pose and clothing appearance we condition both the texture and geometry generators with attribute labels such as clothing types for the geometry and clothing colors for the texture generator. We automatically generated these conditioning labels for the 2D images based on the visual question-answering model BLIP and CLIP. We validate our method on the SCULPT dataset and compare to state-of-the-art 3D generative models for clothed human bodies. Our code and data can be found at https://sculpt.is.tue.mpg.de.

Despite their exceptional generative abilities large T2I diffusion models much like skilled but careless artists often struggle with accurately depicting visual relationships between objects. This issue as we uncover through careful analysis arises from a misaligned text encoder that struggles to interpret specific relationships and differentiate the logical order of associated objects. To resolve this we introduce a novel task termed Relation Rectification aiming to refine the model to accurately represent a given relationship it initially fails to generate. To address this we propose an innovative solution utilizing a Heterogeneous Graph Convolutional Network (HGCN). It models the directional relationships between relation terms and corresponding objects within the input prompts. Specifically we optimize the HGCN on a pair of prompts with identical relational words but reversed object orders supplemented by a few reference images. The lightweight HGCN adjusts the text embeddings generated by the text encoder ensuring accurate reflection of the textual relation in the embedding space. Crucially our method retains the parameters of the text encoder and diffusion model preserving the model's robust performance on unrelated descriptions. We validated our approach on a newly curated dataset of diverse relational data demonstrating both quantitative and qualitative enhancements in generating images with precise visual relations. Project page: https://wuyinwei-hah.github.io/rrnet.github.io/ .

Steganography is the art of hiding secret data into the cover media for covert communication. In recent years more and more deep neural network (DNN)-based steganographic schemes are proposed to train steganographic networks for secret embedding and recovery which are shown to be promising. Compared with the handcrafted steganographic tools steganographic networks tend to be large in size. It raises concerns on how to imperceptibly and effectively transmit these networks to the sender and receiver to facilitate the covert communication. To address this issue we propose in this paper a Purified and Unified Steganographic Network (PUSNet). It performs an ordinary machine learning task in a purified network which could be triggered into steganographic networks for secret embedding or recovery using different keys. We formulate the construction of the PUSNet into a sparse weight filling problem to flexibly switch between the purified and steganographic networks. We further instantiate our PUSNet as an image denoising network with two steganographic networks concealed for secret image embedding and recovery. Comprehensive experiments demonstrate that our PUSNet achieves good performance on secret image embedding secret image recovery and image denoising in a single architecture. It is also shown to be capable of imperceptibly carrying the steganographic networks in a purified network. Steganography is the art of hiding secret data into the cover media for covert communication. In recent years more and more deep neural network (DNN)-based steganographic schemes are proposed to train steganographic networks for secret embedding and recovery which are shown to be promising. Compared with the handcrafted steganographic tools steganographic networks tend to be large in size. It raises concerns on how to imperceptibly and effectively transmit these networks to the sender and receiver to facilitate the covert communication. To address this issue we propose in this paper a Purified and Unified Steganographic Network (PUSNet). It performs an ordinary machine learning task in a purified network which could be triggered into steganographic networks for secret embedding or recovery using different keys. We formulate the construction of the PUSNet into a sparse weight filling problem to flexibly switch between the purified and steganographic networks. We further instantiate our PUSNet as an image denoising network with two steganographic networks concealed for secret image embedding and recovery. Comprehensive experiments demonstrate that our PUSNet achieves good performance on secret image embedding secret image recovery and image denoising in a single architecture. It is also shown to be capable of imperceptibly carrying the steganographic networks in a purified network. Code is available at https://github.com/albblgb/PUSNet