Search

Diffusion probabilistic models have shown significant progress in video generation; however, their computational efficiency is limited by the large number of sampling steps required. Reducing sampling steps often compromises video quality or generation diversity. In this work, we introduce a distillation method that combines variational score distillation and consistency distillation to achieve few-step video generation, maintaining both high quality and diversity. We also propose a latent reward model fine-tuning approach to further enhance video generation performance according to any specified reward metric. This approach reduces memory usage and does not require the reward to be differentiable. Our method demonstrates state-of-the-art performance in few-step generation for 10-second videos (128 frames at 12 FPS). The distilled student model achieves a score of 82.57 on VBench, surpassing the teacher model as well as baseline models Gen-3, T2V-Turbo, and Kling. One-step distillation accelerates the teacher model's diffusion sampling by up to 278.6 times, enabling near real-time generation. Human evaluations further validate the superior performance of our 4-step student models compared to teacher model using 50-step DDIM sampling.

Generating realistic human videos remains a challenging task, with the most effective methods currently relying on a human motion sequence as a control signal. Existing approaches often use existing motion extracted from other videos, which restricts applications to specific motion types and global scene matching. We propose Move-in-2D, a novel approach to generate human motion sequences conditioned on a scene image, allowing for diverse motion that adapts to different scenes. Our approach utilizes a diffusion model that accepts both a scene image and text prompt as inputs, producing a motion sequence tailored to the scene. To train this model, we collect a large-scale video dataset featuring single-human activities, annotating each video with the corresponding human motion as the target output. Experiments demonstrate that our method effectively predicts human motion that aligns with the scene image after projection. Furthermore, we show that the generated motion sequence improves human motion quality in video synthesis tasks., Comment: Project page: https://hhsinping.github.io/Move-in-2D/

Current frontier video diffusion models have demonstrated remarkable results at generating high-quality videos. However, they can only generate short video clips, normally around 10 seconds or 240 frames, due to computation limitations during training. In this work, we show that existing models can be naturally extended to autoregressive video diffusion models without changing the architectures. Our key idea is to assign the latent frames with progressively increasing noise levels rather than a single noise level, which allows for fine-grained condition among the latents and large overlaps between the attention windows. Such progressive video denoising allows our models to autoregressively generate video frames without quality degradation or abrupt scene changes. We present state-of-the-art results on long video generation at 1 minute (1440 frames at 24 FPS). Videos from this paper are available at https://desaixie.github.io/pa-vdm/., Comment: 15 pages, 5 figures. Our video results and code are available at https://desaixie.github.io/pa-vdm/

We present a method to create diffusion-based video models from pretrained Text-to-Image (T2I) models. Recently, AnimateDiff proposed freezing the T2I model while only training temporal layers. We advance this method by proposing a unique architecture, incorporating a mapping network and frame-wise tokens, tailored for video generation while maintaining the diversity and creativity of the original T2I model. Key innovations include novel loss functions for temporal smoothness and a mitigating gradient sampling technique, ensuring realistic and temporally consistent video generation despite limited public video data. We have successfully integrated video-specific inductive biases into the architecture and loss functions. Our method, built on the frozen StableDiffusion model, simplifies training processes and allows for seamless integration with off-the-shelf models like ControlNet and DreamBooth. project page: https://kwonminki.github.io/HARIVO, Comment: ECCV2024

While AI-generated content has garnered significant attention, achieving photo-realistic video synthesis remains a formidable challenge. Despite the promising advances in diffusion models for video generation quality, the complex model architecture and substantial computational demands for both training and inference create a significant gap between these models and real-world applications. This paper presents SNED, a superposition network architecture search method for efficient video diffusion model. Our method employs a supernet training paradigm that targets various model cost and resolution options using a weight-sharing method. Moreover, we propose the supernet training sampling warm-up for fast training optimization. To showcase the flexibility of our method, we conduct experiments involving both pixel-space and latent-space video diffusion models. The results demonstrate that our framework consistently produces comparable results across different model options with high efficiency. According to the experiment for the pixel-space video diffusion model, we can achieve consistent video generation results simultaneously across 64 x 64 to 256 x 256 resolutions with a large range of model sizes from 640M to 1.6B number of parameters for pixel-space video diffusion models., Comment: Accepted in CVPR 2024

The success of denoising diffusion models in representing rich data distributions over 2D raster images has prompted research on extending them to other data representations, such as vector graphics. Unfortunately due to their variable structure and scarcity of vector training data, directly applying diffusion models on this domain remains a challenging problem. Using workarounds like optimization via Score Distillation Sampling (SDS) is also fraught with difficulty, as vector representations are non trivial to directly optimize and tend to result in implausible geometries such as redundant or self-intersecting shapes. NIVeL addresses these challenges by reinterpreting the problem on an alternative, intermediate domain which preserves the desirable properties of vector graphics -- mainly sparsity of representation and resolution-independence. This alternative domain is based on neural implicit fields expressed in a set of decomposable, editable layers. Based on our experiments, NIVeL produces text-to-vector graphics results of significantly better quality than the state-of-the-art.

Diffusion Models (DMs) have exhibited superior performance in generating high-quality and diverse images. However, this exceptional performance comes at the cost of expensive architectural design, particularly due to the attention module heavily used in leading models. Existing works mainly adopt a retraining process to enhance DM efficiency. This is computationally expensive and not very scalable. To this end, we introduce the Attention-driven Training-free Efficient Diffusion Model (AT-EDM) framework that leverages attention maps to perform run-time pruning of redundant tokens, without the need for any retraining. Specifically, for single-denoising-step pruning, we develop a novel ranking algorithm, Generalized Weighted Page Rank (G-WPR), to identify redundant tokens, and a similarity-based recovery method to restore tokens for the convolution operation. In addition, we propose a Denoising-Steps-Aware Pruning (DSAP) approach to adjust the pruning budget across different denoising timesteps for better generation quality. Extensive evaluations show that AT-EDM performs favorably against prior art in terms of efficiency (e.g., 38.8% FLOPs saving and up to 1.53x speed-up over Stable Diffusion XL) while maintaining nearly the same FID and CLIP scores as the full model. Project webpage: https://atedm.github.io., Comment: Accepted to IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) 2024

Video super-resolution (VSR) approaches have shown impressive temporal consistency in upsampled videos. However, these approaches tend to generate blurrier results than their image counterparts as they are limited in their generative capability. This raises a fundamental question: can we extend the success of a generative image upsampler to the VSR task while preserving the temporal consistency? We introduce VideoGigaGAN, a new generative VSR model that can produce videos with high-frequency details and temporal consistency. VideoGigaGAN builds upon a large-scale image upsampler -- GigaGAN. Simply inflating GigaGAN to a video model by adding temporal modules produces severe temporal flickering. We identify several key issues and propose techniques that significantly improve the temporal consistency of upsampled videos. Our experiments show that, unlike previous VSR methods, VideoGigaGAN generates temporally consistent videos with more fine-grained appearance details. We validate the effectiveness of VideoGigaGAN by comparing it with state-of-the-art VSR models on public datasets and showcasing video results with $8\times$ super-resolution., Comment: project page: https://videogigagan.github.io/

Image customization has been extensively studied in text-to-image (T2I) diffusion models, leading to impressive outcomes and applications. With the emergence of text-to-video (T2V) diffusion models, its temporal counterpart, motion customization, has not yet been well investigated. To address the challenge of one-shot video motion customization, we propose Customize-A-Video that models the motion from a single reference video and adapts it to new subjects and scenes with both spatial and temporal varieties. It leverages low-rank adaptation (LoRA) on temporal attention layers to tailor the pre-trained T2V diffusion model for specific motion modeling. To disentangle the spatial and temporal information during training, we introduce a novel concept of appearance absorbers that detach the original appearance from the reference video prior to motion learning. The proposed modules are trained in a staged pipeline and inferred in a plug-and-play fashion, enabling easy extensions to various downstream tasks such as custom video generation and editing, video appearance customization and multiple motion combination. Our project page can be found at https://customize-a-video.github.io., Comment: Accepted by ECCV 2024. Project page: https://customize-a-video.github.io

We present VecFusion, a new neural architecture that can generate vector fonts with varying topological structures and precise control point positions. Our approach is a cascaded diffusion model which consists of a raster diffusion model followed by a vector diffusion model. The raster model generates low-resolution, rasterized fonts with auxiliary control point information, capturing the global style and shape of the font, while the vector model synthesizes vector fonts conditioned on the low-resolution raster fonts from the first stage. To synthesize long and complex curves, our vector diffusion model uses a transformer architecture and a novel vector representation that enables the modeling of diverse vector geometry and the precise prediction of control points. Our experiments show that, in contrast to previous generative models for vector graphics, our new cascaded vector diffusion model generates higher quality vector fonts, with complex structures and diverse styles.

We propose the first Large Reconstruction Model (LRM) that predicts the 3D model of an object from a single input image within just 5 seconds. In contrast to many previous methods that are trained on small-scale datasets such as ShapeNet in a category-specific fashion, LRM adopts a highly scalable transformer-based architecture with 500 million learnable parameters to directly predict a neural radiance field (NeRF) from the input image. We train our model in an end-to-end manner on massive multi-view data containing around 1 million objects, including both synthetic renderings from Objaverse and real captures from MVImgNet. This combination of a high-capacity model and large-scale training data empowers our model to be highly generalizable and produce high-quality 3D reconstructions from various testing inputs, including real-world in-the-wild captures and images created by generative models. Video demos and interactable 3D meshes can be found on our LRM project webpage: https://yiconghong.me/LRM., Comment: ICLR 2024

We present ASSET, a neural architecture for automatically modifying an input high-resolution image according to a user's edits on its semantic segmentation map. Our architecture is based on a transformer with a novel attention mechanism. Our key idea is to sparsify the transformer's attention matrix at high resolutions, guided by dense attention extracted at lower image resolutions. While previous attention mechanisms are computationally too expensive for handling high-resolution images or are overly constrained within specific image regions hampering long-range interactions, our novel attention mechanism is both computationally efficient and effective. Our sparsified attention mechanism is able to capture long-range interactions and context, leading to synthesizing interesting phenomena in scenes, such as reflections of landscapes onto water or flora consistent with the rest of the landscape, that were not possible to generate reliably with previous convnets and transformer approaches. We present qualitative and quantitative results, along with user studies, demonstrating the effectiveness of our method., Comment: SIGGRAPH 2022 - Journal Track

This paper introduces a model for producing stylized line drawings from 3D shapes. The model takes a 3D shape and a viewpoint as input, and outputs a drawing with textured strokes, with variations in stroke thickness, deformation, and color learned from an artist's style. The model is fully differentiable. We train its parameters from a single training drawing of another 3D shape. We show that, in contrast to previous image-based methods, the use of a geometric representation of 3D shape and 2D strokes allows the model to transfer important aspects of shape and texture style while preserving contours. Our method outputs the resulting drawing in a vector representation, enabling richer downstream analysis or editing in interactive applications., Comment: Accepted to ICCV 2021

We propose a novel, end-to-end trainable, deep network called ParSeNet that decomposes a 3D point cloud into parametric surface patches, including B-spline patches as well as basic geometric primitives. ParSeNet is trained on a large-scale dataset of man-made 3D shapes and captures high-level semantic priors for shape decomposition. It handles a much richer class of primitives than prior work, and allows us to represent surfaces with higher fidelity. It also produces repeatable and robust parametrizations of a surface compared to purely geometric approaches. We present extensive experiments to validate our approach against analytical and learning-based alternatives. Our source code is publicly available at: https://hippogriff.github.io/parsenet.

This paper introduces a method for learning to generate line drawings from 3D models. Our architecture incorporates a differentiable module operating on geometric features of the 3D model, and an image-based module operating on view-based shape representations. At test time, geometric and view-based reasoning are combined with the help of a neural module to create a line drawing. The model is trained on a large number of crowdsourced comparisons of line drawings. Experiments demonstrate that our method achieves significant improvements in line drawing over the state-of-the-art when evaluated on standard benchmarks, resulting in drawings that are comparable to those produced by experienced human artists., Comment: Accepted to CVPR 2020

Constructive Solid Geometry (CSG) is a geometric modeling technique that defines complex shapes by recursively applying boolean operations on primitives such as spheres and cylinders. We present CSGNe, a deep network architecture that takes as input a 2D or 3D shape and outputs a CSG program that models it. Parsing shapes into CSG programs is desirable as it yields a compact and interpretable generative model. However, the task is challenging since the space of primitives and their combinations can be prohibitively large. CSGNe uses a convolutional encoder and recurrent decoder based on deep networks to map shapes to modeling instructions in a feed-forward manner and is significantly faster than bottom-up approaches. We investigate two architectures for this task --- a vanilla encoder (CNN) - decoder (RNN) and another architecture that augments the encoder with an explicit memory module based on the program execution stack. The stack augmentation improves the reconstruction quality of the generated shape and learning efficiency. Our approach is also more effective as a shape primitive detector compared to a state-of-the-art object detector. Finally, we demonstrate CSGNet can be trained on novel datasets without program annotations through policy gradient techniques., Comment: arXiv admin note: substantial text overlap with arXiv:1712.08290

Object functionality is often expressed through part articulation -- as when the two rigid parts of a scissor pivot against each other to perform the cutting function. Such articulations are often similar across objects within the same functional category. In this paper, we explore how the observation of different articulation states provides evidence for part structure and motion of 3D objects. Our method takes as input a pair of unsegmented shapes representing two different articulation states of two functionally related objects, and induces their common parts along with their underlying rigid motion. This is a challenging setting, as we assume no prior shape structure, no prior shape category information, no consistent shape orientation, the articulation states may belong to objects of different geometry, plus we allow inputs to be noisy and partial scans, or point clouds lifted from RGB images. Our method learns a neural network architecture with three modules that respectively propose correspondences, estimate 3D deformation flows, and perform segmentation. To achieve optimal performance, our architecture alternates between correspondence, deformation flow, and segmentation prediction iteratively in an ICP-like fashion. Our results demonstrate that our method significantly outperforms state-of-the-art techniques in the task of discovering articulated parts of objects. In addition, our part induction is object-class agnostic and successfully generalizes to new and unseen objects.

We present a neural architecture that takes as input a 2D or 3D shape and outputs a program that generates the shape. The instructions in our program are based on constructive solid geometry principles, i.e., a set of boolean operations on shape primitives defined recursively. Bottom-up techniques for this shape parsing task rely on primitive detection and are inherently slow since the search space over possible primitive combinations is large. In contrast, our model uses a recurrent neural network that parses the input shape in a top-down manner, which is significantly faster and yields a compact and easy-to-interpret sequence of modeling instructions. Our model is also more effective as a shape detector compared to existing state-of-the-art detection techniques. We finally demonstrate that our network can be trained on novel datasets without ground-truth program annotations through policy gradient techniques., Comment: Accepted at CVPR-2018