CVPR 2024 Papers

Recently there has been an increased interest in the practical problem of learning multiple dense scene understanding tasks from partially annotated data where each training sample is only labeled for a subset of the tasks. The missing of task labels in training leads to low-quality and noisy predictions as can be observed from state-of-the-art methods. To tackle this issue we reformulate the partially-labeled multi-task dense prediction as a pixel-level denoising problem and propose a novel multi-task denoising diffusion framework coined as DiffusionMTL. It designs a joint diffusion and denoising paradigm to model a potential noisy distribution in the task prediction or feature maps and generate rectified outputs for different tasks. To exploit multi-task consistency in denoising we further introduce a Multi-Task Conditioning strategy which can implicitly utilize the complementary nature of the tasks to help learn the unlabeled tasks leading to an improvement in the denoising performance of the different tasks. Extensive quantitative and qualitative experiments demonstrate that the proposed multi-task denoising diffusion model can significantly improve multi-task prediction maps and outperform the state-of-the-art methods on three challenging multi-task benchmarks under two different partial-labeling evaluation settings. The code is available at https://prismformore.github.io/diffusionmtl/.

We present a simple yet effective technique to estimate lighting in a single input image. Current techniques rely heavily on HDR panorama datasets to train neural networks to regress an input with limited field-of-view to a full environment map. However these approaches often struggle with real-world uncontrolled settings due to the limited diversity and size of their datasets. To address this problem we leverage diffusion models trained on billions of standard images to render a chrome ball into the input image. Despite its simplicity this task remains challenging: the diffusion models often insert incorrect or inconsistent objects and cannot readily generate chrome balls in HDR format. Our research uncovers a surprising relationship between the appearance of chrome balls and the initial diffusion noise map which we utilize to consistently generate high-quality chrome balls. We further fine-tune an LDR diffusion model (Stable Diffusion XL) with LoRA enabling it to perform exposure bracketing for HDR light estimation. Our method produces convincing light estimates across diverse settings and demonstrates superior generalization to in-the-wild scenarios.

Text-guided domain adaptation and generation of 3D-aware portraits find many applications in various fields. However due to the lack of training data and the challenges in handling the high variety of geometry and appearance the existing methods for these tasks suffer from issues like inflexibility instability and low fidelity. In this paper we propose a novel framework DiffusionGAN3D which boosts text-guided 3D domain adaptation and generation by combining 3D GANs and diffusion priors. Specifically we integrate the pre-trained 3D generative models (e.g. EG3D) and text-to-image diffusion models. The former provides a strong foundation for stable and high-quality avatar generation from text. And the diffusion models in turn offer powerful priors and guide the 3D generator finetuning with informative direction to achieve flexible and efficient text-guided domain adaptation. To enhance the diversity in domain adaptation and the generation capability in text-to-avatar we introduce the relative distance loss and case-specific learnable triplane respectively. Besides we design a progressive texture refinement module to improve the texture quality for both tasks above. Extensive experiments demonstrate that the proposed framework achieves excellent results in both domain adaptation and text-to-avatar tasks outperforming existing methods in terms of generation quality and efficiency. The project homepage is at https://younglbw.github.io/DiffusionGAN3D-homepage/.

DiffusionAvatars synthesizes a high-fidelity 3D head avatar of a person offering intuitive control over both pose and expression. We propose a diffusion-based neural renderer that leverages generic 2D priors to produce compelling images of faces. For coarse guidance of the expression and head pose we render a neural parametric head model (NPHM) from the target viewpoint which acts as a proxy geometry of the person. Additionally to enhance the modeling of intricate facial expressions we condition DiffusionAvatars directly on the expression codes obtained from NPHM via cross-attention. Finally to synthesize consistent surface details across different viewpoints and expressions we rig learnable spatial features to the head's surface via TriPlane lookup in NPHM's canonical space. We train DiffusionAvatars on RGB videos and corresponding fitted NPHM meshes of a person and test the obtained avatars in both self-reenactment and animation scenarios. Our experiments demonstrate that DiffusionAvatars generates temporally consistent and visually appealing videos for novel poses and expressions of a person outperforming existing approaches.

We present a new multi-modal face image generation method that converts a text prompt and a visual input such as a semantic mask or scribble map into a photo-realistic face image. To do this we combine the strengths of Generative Adversarial networks (GANs) and diffusion models (DMs) by employing the multi-modal features in the DM into the latent space of the pre-trained GANs. We present a simple mapping and a style modulation network to link two models and convert meaningful representations in feature maps and attention maps into latent codes. With GAN inversion the estimated latent codes can be used to generate 2D or 3D-aware facial images. We further present a multi-step training strategy that reflects textual and structural representations into the generated image. Our proposed network produces realistic 2D multi-view and stylized face images which align well with inputs. We validate our method by using pre-trained 2D and 3D GANs and our results outperform existing methods. Our project page is available at https://github.com/1211sh/Diffusiondriven_GAN-Inversion/.

Recovering degraded low-resolution text images is challenging especially for Chinese text images with complex strokes and severe degradation in real-world scenarios. Ensuring both text fidelity and style realness is crucial for high-quality text image super-resolution. Recently diffusion models have achieved great success in natural image synthesis and restoration due to their powerful data distribution modeling abilities and data generation capabilities. In this work we propose an Image Diffusion Model (IDM) to restore text images with realistic styles. For diffusion models they are not only suitable for modeling realistic image distribution but also appropriate for learning text distribution. Since text prior is important to guarantee the correctness of the restored text structure according to existing arts we also propose a Text Diffusion Model (TDM) for text recognition which can guide IDM to generate text images with correct structures. We further propose a Mixture of Multi-modality module (MoM) to make these two diffusion models cooperate with each other in all the diffusion steps. Extensive experiments on synthetic and real-world datasets demonstrate that our Diffusion-based Blind Text Image Super-Resolution (DiffTSR) can restore text images with more accurate text structures as well as more realistic appearances simultaneously. Code is available at https://github.com/YuzheZhang-1999/DiffTSR.

Reconstructing a clothed human from a single-view image has several challenging issues including flexibly representing various body shapes and poses estimating complete 3D geometry and consistent texture and achieving more fine-grained details. To address them we propose a new diffusion-based Fourier occupancy field method to improve the human representing ability and the geometry generating ability. First we estimate the back-view image from the given reference image by incorporating a style consistency constraint. Then we extract multi-scale features of the two images as conditional and design a diffusion model to generate the Fourier occupancy field in the wavelet domain. We refine the initial estimated Fourier occupancy field with image features as conditions to improve the geometric accuracy. Finally the reference and estimated back-view images are mapped onto the human model creating a textured clothed human model. Substantial experiments are conducted and the experimental results show that our method outperforms the state-of-the-art methods in geometry and texture reconstruction performance.

Diffusion models excel at modeling complex and multimodal trajectory distributions for decision-making and control. Reward-gradient guided denoising has been recently proposed to generate trajectories that maximize both a differentiable reward function and the likelihood under the data distribution captured by a diffusion model. Reward-gradient guided denoising requires a differentiable reward function fitted to both clean and noised samples limiting its applicability as a general trajectory optimizer. In this paper we propose DiffusionES a method that combines gradient-free optimization with trajectory denoising to optimize black-box non-differentiable objectives while staying in the data manifold. Diffusion-ES samples trajectories during evolutionary search from a diffusion model and scores them using a black-box reward function. It mutates high-scoring trajectories using a truncated diffusion process that applies a small number of noising and denoising steps allowing for much more efficient exploration of the solution space. We show that DiffusionES achieves state-of-the-art performance on nuPlan an established closed-loop planning benchmark for autonomous driving. Diffusion-ES outperforms existing sampling-based planners reactive deterministic or diffusion-based policies and reward-gradient guidance. Additionally we show that unlike prior guidance methods our method can optimize non-differentiable language-shaped reward functions generated by few-shot LLM prompting. When guided by a human teacher that issues instructions to follow our method can generate novel highly complex behaviors such as aggressive lane weaving which are not present in the training data. This allows us to solve the hardest nuPlan scenarios which are beyond the capabilities of existing trajectory optimization methods and driving policies.

Diffusion generative modeling has become a promising approach for learning robotic manipulation tasks from stochastic human demonstrations. In this paper we present Diffusion-EDFs a novel SE(3)-equivariant diffusion-based approach for visual robotic manipulation tasks. We show that our proposed method achieves remarkable data efficiency requiring only 5 to 10 human demonstrations for effective end-to-end training in less than an hour. Furthermore our benchmark experiments demonstrate that our approach has superior generalizability and robustness compared to state-of-the-art methods. Lastly we validate our methods with real hardware experiments.

Score distillation sampling (SDS) has been widely adopted to overcome the absence of unseen views in reconstructing 3D objects from a single image. It leverages pre-trained 2D diffusion models as teacher to guide the reconstruction of student 3D models. Despite their remarkable success SDS-based methods often encounter geometric artifacts and texture saturation. We find out the crux is the overlooked indiscriminate treatment of diffusion time-steps during optimization: it unreasonably treats the student-teacher knowledge distillation to be equal at all time-steps and thus entangles coarse-grained and fine-grained modeling. Therefore we propose the Diffusion Time-step Curriculum one-image-to-3D pipeline (DTC123) which involves both the teacher and student models collaborating with the time-step curriculum in a coarse-to-fine manner. Extensive experiments on NeRF4 RealFusion15 GSO and Level50 benchmark demonstrate that DTC123 can produce multi-view consistent high-quality and diverse 3D assets. Codes and more generation demos will be released in https://github.com/yxymessi/DTC123.

Reflectance bounds the frequency spectrum of illumination in the object appearance. In this paper we introduce the first stochastic inverse rendering method which recovers the attenuated frequency spectrum of an illumination jointly with the reflectance of an object of known geometry from a single image. Our key idea is to solve this blind inverse problem in the reflectance map an appearance representation invariant to the underlying geometry by learning to reverse the image formation with a novel diffusion model which we refer to as the Diffusion Reflectance Map Network (DRMNet). Given an observed reflectance map converted and completed from the single input image DRMNet generates a reflectance map corresponding to a perfect mirror sphere while jointly estimating the reflectance. The forward process can be understood as gradually filtering a natural illumination with lower and lower frequency reflectance and additive Gaussian noise. DRMNet learns to invert this process with two subnetworks IllNet and RefNet which work in concert towards this joint estimation. The network is trained on an extensive synthetic dataset and is demonstrated to generalize to real images showing state-of-the-art accuracy on established datasets.

In recent advancements in high-fidelity image generation Denoising Diffusion Probabilistic Models (DDPMs) have emerged as a key player. However their application at high resolutions presents significant computational challenges. Current methods such as patchifying expedite processes in UNet and Transformer architectures but at the expense of representational capacity. Addressing this we introduce the Diffusion State Space Model (DiffuSSM) an architecture that supplants attention mechanisms with a more scalable state space model backbone. This approach effectively handles higher resolutions without resorting to global compression thus preserving detailed image representation throughout the diffusion process. Our focus on FLOP-efficient architectures in diffusion training marks a significant step forward. Comprehensive evaluations on both ImageNet and LSUN datasets at two resolutions demonstrate that DiffuSSMs are on par or even outperform existing diffusion models with attention modules in FID and Inception Score metrics while significantly reducing total FLOP usage.

Large language models (LLMs) are fine-tuned using human comparison data with Reinforcement Learning from Human Feedback (RLHF) methods to make them better aligned with users' preferences. In contrast to LLMs human preference learning has not been widely explored in text-to-image diffusion models; the best existing approach is to fine-tune a pretrained model using carefully curated high quality images and captions to improve visual appeal and text alignment. We propose Diffusion-DPO a method to align diffusion models to human preferences by directly optimizing on human comparison data. Diffusion-DPO is adapted from the recently developed Direct Preference Optimization (DPO) a simpler alternative to RLHF which directly optimizes a policy that best satisfies human preferences under a classification objective. We re-formulate DPO to account for a diffusion model notion of likelihood utilizing the evidence lower bound to derive a differentiable objective. Using the Pick-a-pic dataset of 851K crowdsourced pairwise preferences we fine-tune the base model of the state-of-the-art Stable Diffusion XL (SDXL)-1.0 model with Diffusion-DPO. Our fine-tuned base model significantly outperforms both base SDXL-1.0 and the larger SDXL-1.0 model consisting of an additional refinement model in human evaluation improving visual appeal and prompt alignment. We also develop a variant that uses AI feedback and has comparable performance to training on human preferences opening the door for scaling of diffusion model alignment methods.

Diffusion handles is a novel approach to enable 3D object edits on diffusion images requiring only existing pre-trained diffusion models depth estimation without any fine-tuning or 3D object retrieval. The edited results remain plausible photo-real and preserve object identity. Diffusion handles address a critically missing facet of generative image-based creative design. Our key insight is to lift diffusion activations for a selected object to 3D using a proxy depth 3D-transform the depth and associated activations and project them back to image space. The diffusion process guided by the manipulated activations produces plausible edited images showing complex 3D occlusion and lighting effects. We evaluate diffusion handles: quantitatively on a large synthetic data benchmark; and qualitatively by a user study showing our output to be more plausible and better than prior art at both 3D editing and identity control.

We present Diff3F as a simple robust and class-agnostic feature descriptor that can be computed for untextured input shapes (meshes or point clouds). Our method distills diffusion features from image foundational models onto input shapes. Specifically we use the input shapes to produce depth and normal maps as guidance for conditional image synthesis. In the process we produce (diffusion) features in 2D that we subsequently lift and aggregate on the original surface. Our key observation is that even if the conditional image generations obtained from multi-view rendering of the input shapes are inconsistent the associated image features are robust and hence can be directly aggregated across views. This produces semantic features on the input shapes without requiring additional data or training. We perform extensive experiments on multiple benchmarks (SHREC'19 SHREC'20 FAUST and TOSCA) and demonstrate that our features being semantic instead of geometric produce reliable correspondence across both isometric and non-isometrically related shape families. Code is available via the project webpage at https://diff3f.github.io/

Recently a number of image-mixing-based augmentation techniques have been introduced to improve the generalization of deep neural networks. In these techniques two or more randomly selected natural images are mixed together to generate an augmented image. Such methods may not only omit important portions of the input images but also introduce label ambiguities by mixing images across labels resulting in misleading supervisory signals. To address these limitations we propose DIFFUSEMIX a novel data augmentation technique that leverages a diffusion model to reshape training images supervised by our bespoke conditional prompts. First concatenation of a partial natural image and its generated counterpart is obtained which helps in avoiding the generation of unrealistic images or label ambiguities. Then to enhance resilience against adversarial attacks and improves safety measures a randomly selected structural pattern from a set of fractal images is blended into the concatenated image to form the final augmented image for training. Our empirical results on seven different datasets reveal that DIFFUSEMIX achieves superior performance compared to existing state- of-the-art methods on tasks including general classification fine-grained classification fine-tuning data scarcity and adversarial robustness.

Producing quality segmentation masks for images is a fundamental problem in computer vision. Recent research has explored large-scale supervised training to enable zero-shot transfer segmentation on virtually any image style and unsupervised training to enable segmentation without dense annotations. However constructing a model capable of segmenting anything in a zero-shot manner without any annotations is still challenging. In this paper we propose to utilize the self-attention layers in stable diffusion models to achieve this goal because the pre-trained stable diffusion model has learned inherent concepts of objects within its attention layers. Specifically we introduce a simple yet effective iterative merging process based on measuring KL divergence among attention maps to merge them into valid segmentation masks. The proposed method does not require any training or language dependency to extract quality segmentation for any images. On COCO-Stuff-27 our method surpasses the prior unsupervised zero-shot transfer SOTA method by an absolute 26% in pixel accuracy and 17% in mean IoU.

We present DiffuScene for indoor 3D scene synthesis based on a novel scene configuration denoising diffusion model. It generates 3D instance properties stored in an unordered object set and retrieves the most similar geometry for each object configuration which is characterized as a concatenation of different attributes including location size orientation semantics and geometry features. We introduce a diffusion network to synthesize a collection of 3D indoor objects by denoising a set of unordered object attributes. Unordered parametrization simplifies and eases the joint distribution approximation. The shape feature diffusion facilitates natural object placements including symmetries. Our method enables many downstream applications including scene completion scene arrangement and text-conditioned scene synthesis. Experiments on the 3D-FRONT dataset show that our method can synthesize more physically plausible and diverse indoor scenes than state-of-the-art methods. Extensive ablation studies verify the effectiveness of our design choice in scene diffusion models.

Audio-visual saliency prediction can draw support from diverse modality complements but further performance enhancement is still challenged by customized architectures as well as task-specific loss functions. In recent studies denoising diffusion models have shown more promising in unifying task frameworks owing to their inherent ability of generalization. Following this motivation a novel Diffusion architecture for generalized audio-visual Saliency prediction (DiffSal) is proposed in this work which formulates the prediction problem as a conditional generative task of the saliency map by utilizing input audio and video as the conditions. Based on the spatio-temporal audio-visual features an extra network Saliency-UNet is designed to perform multi-modal attention modulation for progressive refinement of the ground-truth saliency map from the noisy map. Extensive experiments demonstrate that the proposed DiffSal can achieve excellent performance across six challenging audio-visual benchmarks with an average relative improvement of 6.3% over the previous state-of-the-art results by six metrics.