VideoMamba、Video Reframing、DiffuMatting、VideoEdit、Style2Talker

本文首发于公众号：机器感知

VideoMamba、Video Reframing、DiffuMatting、VideoEdit、Style2Talker

 

Audio-Synchronized Visual Animation

Current visual generation methods can produce high quality videos guided by texts. However, effectively controlling object dynamics remains a challenge. This work explores audio as a cue to generate temporally synchronized image animations. We introduce Audio Synchronized Visual Animation (ASVA), a task animating a static image to demonstrate motion dynamics, temporally guided by audio clips across multiple classes. To this end, we present AVSync15, a dataset curated from VGGSound with videos featuring synchronized audio visual events across 15 categories. We also present a diffusion model, AVSyncD, capable of generating dynamic animations guided by audios. Extensive evaluations validate AVSync15 as a reliable benchmark for synchronized generation and demonstrate our models superior performance. We further explore AVSyncDs potential in a variety of audio synchronized generation tasks, from generating full videos without a base image to controlling object motions with various sounds. We hope our established benchmark can open new avenues for controllable visual generation.

Adaptive Multi-modal Fusion of Spatially Variant Kernel Refinement with Diffusion Model for Blind Image Super-Resolution

Pre-trained diffusion models utilized for image generation encapsulate a substantial reservoir of a priori knowledge pertaining to intricate textures. Harnessing the potential of leveraging this a priori knowledge in the context of image super-resolution presents a compelling avenue. Nonetheless, prevailing diffusion-based methodologies presently overlook the constraints imposed by degradation information on the diffusion process. Furthermore, these methods fail to consider the spatial variability inherent in the estimated blur kernel, stemming from factors such as motion jitter and out-of-focus elements in open-environment scenarios. This oversight results in a notable deviation of the image super-resolution effect from fundamental realities. Finally, we introduce the Adaptive Multi-Modal Fusion (AMF) module to align the information from three modalities: low-resolution images, depth maps, and blur kernels. This alignment can constrain the diffusion model to generate more authentic SR results. Quantitative and qualitative experiments affirm the superiority of our approach, while ablation experiments corroborate the effectiveness of the modules we have proposed.

Enhancing Expressiveness in Dance Generation via Integrating Frequency and Music Style Information

Dance generation, as a branch of human motion generation, has attracted increasing attention. Recently, a few works attempt to enhance dance expressiveness, which includes genre matching, beat alignment, and dance dynamics, from certain aspects. However, the enhancement is quite limited as they lack comprehensive consideration of the aforementioned three factors. In this paper, we propose ExpressiveBailando, a novel dance generation method designed to generate expressive dances, concurrently taking all three factors into account. Specifically, we mitigate the issue of speed homogenization by incorporating frequency information into VQ-VAE, thus improving dance dynamics. Additionally, we integrate music style information by extracting genre- and beat-related features with a pre-trained music model, hence achieving improvements in the other two factors. Extensive experimental results demonstrate that our proposed method can generate dances with high expressiveness and outperforms existing methods both qualitatively and quantitatively.

Reframe Anything: LLM Agent for Open World Video Reframing

The proliferation of mobile devices and social media has revolutionized content dissemination, with short-form video becoming increasingly prevalent. This shift has introduced the challenge of video reframing to fit various screen aspect ratios, a process that highlights the most compelling parts of a video. Traditionally, video reframing is a manual, time-consuming task requiring professional expertise, which incurs high production costs. A potential solution is to adopt some machine learning models, such as video salient object detection, to automate the process. However, these methods often lack generalizability due to their reliance on specific training data. The advent of powerful large language models (LLMs) open new avenues for AI capabilities. Building on this, we introduce Reframe Any Video Agent (RAVA), a LLM-based agent that leverages visual foundation models and human instructions to restructure visual content for video reframing. RAVA operates in three stages: perception, where it interprets user instructions and video content; planning, where it determines aspect ratios and reframing strategies; and execution, where it invokes the editing tools to produce the final video.

VidProM: A Million-scale Real Prompt-Gallery Dataset for Text-to-Video Diffusion Models

The arrival of Sora marks a new era for text-to-video diffusion models, bringing significant advancements in video generation and potential applications. However, Sora, as well as other text-to-video diffusion models, highly relies on the prompts, and there is no publicly available dataset featuring a study of text-to-video prompts. In this paper, we introduce VidProM, the first large-scale dataset comprising 1.67 million unique text-to-video prompts from real users. Additionally, the dataset includes 6.69 million videos generated by four state-of-the-art diffusion models and some related data. We initially demonstrate the curation of this large-scale dataset, which is a time-consuming and costly process. Subsequently, we show how the proposed VidProM differs from DiffusionDB, a large-scale prompt-gallery dataset for image generation. Based on the analysis of these prompts, we identify the necessity for a new prompt dataset specifically designed for text-to-video generation and gain insights into the preferences of real users when creating videos. We make the collected dataset VidProM publicly available at GitHub and Hugging Face under the CC-BY- NC 4.0 License.

MACE: Mass Concept Erasure in Diffusion Models

The rapid expansion of large-scale text-to-image diffusion models has raised growing concerns regarding their potential misuse in creating harmful or misleading content. In this paper, we introduce MACE, a finetuning framework for the task of mass concept erasure. This task aims to prevent models from generating images that embody unwanted concepts when prompted. Existing concept erasure methods are typically restricted to handling fewer than five concepts simultaneously and struggle to find a balance between erasing concept synonyms (generality) and maintaining unrelated concepts (specificity). In contrast, MACE differs by successfully scaling the erasure scope up to 100 concepts and by achieving an effective balance between generality and specificity. This is achieved by leveraging closed-form cross-attention refinement along with LoRA finetuning, collectively eliminating the information of undesirable concepts. Furthermore, MACE integrates multiple LoRAs without mutual interference. We conduct extensive evaluations of MACE against prior methods across four different tasks: object erasure, celebrity erasure, explicit content erasure, and artistic style erasure.

DiffuMatting: Synthesizing Arbitrary Objects with Matting-level Annotation

Due to the difficulty and labor-consuming nature of getting highly accurate or matting annotations, there only exists a limited amount of highly accurate labels available to the public. To tackle this challenge, we propose a DiffuMatting which inherits the strong Everything generation ability of diffusion and endows the power of "matting anything". Our DiffuMatting can 1). act as an anything matting factory with high accurate annotations 2). be well-compatible with community LoRAs or various conditional control approaches to achieve the community-friendly art design and controllable generation. Specifically, inspired by green-screen-matting, we aim to teach the diffusion model to paint on a fixed green screen canvas. To this end, a large-scale greenscreen dataset (Green100K) is collected as a training dataset for DiffuMatting. Secondly, a green background control loss is proposed to keep the drawing board as a pure green color to distinguish the foreground and background. To ensure the synthesized object has more edge details, a detailed-enhancement of transition boundary loss is proposed as a guideline to generate objects with more complicated edge structures. Aiming to simultaneously generate the object and its matting annotation, we build a matting head to make a green color removal in the latent space of the VAE decoder. Our DiffuMatting shows several potential applications (e.g., matting-data generator, community-friendly art design and controllable generation). As a matting-data generator, DiffuMatting synthesizes general object and portrait matting sets, effectively reducing the relative MSE error by 15.4% in General Object Matting and 11.4% in Portrait Matting tasks.

MoST: Motion Style Transformer between Diverse Action Contents

While existing motion style transfer methods are effective between two motions with identical content, their performance significantly diminishes when transferring style between motions with different contents. This challenge lies in the lack of clear separation between content and style of a motion. To tackle this challenge, we propose a novel motion style transformer that effectively disentangles style from content and generates a plausible motion with transferred style from a source motion. Our distinctive approach to achieving the goal of disentanglement is twofold: (1) a new architecture for motion style transformer with 'part-attentive style modulator across body parts' and 'Siamese encoders that encode style and content features separately'; (2) style disentanglement loss. Our method outperforms existing methods and demonstrates exceptionally high quality, particularly in motion pairs with different contents, without the need for heuristic post-processing.

FastVideoEdit: Leveraging Consistency Models for Efficient Text-to-Video Editing

Diffusion models have demonstrated remarkable capabilities in text-to-image and text-to-video generation, opening up possibilities for video editing based on textual input. However, the computational cost associated with sequential sampling in diffusion models poses challenges for efficient video editing. Existing approaches relying on image generation models for video editing suffer from time-consuming one-shot fine-tuning, additional condition extraction, or DDIM inversion, making real-time applications impractical. In this work, we propose FastVideoEdit, an efficient zero-shot video editing approach inspired by Consistency Models (CMs). By leveraging the self-consistency property of CMs, we eliminate the need for time-consuming inversion or additional condition extraction, reducing editing time. Our method enables direct mapping from source video to target video with strong preservation ability utilizing a special variance schedule. This results in improved speed advantages, as fewer sampling steps can be used while maintaining comparable generation quality. Experimental results validate the state-of-the-art performance and speed advantages of FastVideoEdit across evaluation metrics encompassing editing speed, temporal consistency, and text-video alignment.

Style2Talker: High-Resolution Talking Head Generation with Emotion Style and Art Style

Although automatically animating audio-driven talking heads has recently received growing interest, previous efforts have mainly concentrated on achieving lip synchronization with the audio, neglecting two crucial elements for generating expressive videos: emotion style and art style. In this paper, we present an innovative audio-driven talking face generation method called Style2Talker. It involves two stylized stages, namely Style-E and Style-A, which integrate text-controlled emotion style and picture-controlled art style into the final output. In order to prepare the scarce emotional text descriptions corresponding to the videos, we propose a labor-free paradigm that employs large-scale pretrained models to automatically annotate emotional text labels for existing audiovisual datasets. Incorporating the synthetic emotion texts, the Style-E stage utilizes a large-scale CLIP model to extract emotion representations, which are combined with the audio, serving as the condition for an efficient latent diffusion model designed to produce emotional motion coefficients of a 3DMM model. Moving on to the Style-A stage, we develop a coefficient-driven motion generator and an art-specific style path embedded in the well-known StyleGAN. This allows us to synthesize high-resolution artistically stylized talking head videos using the generated emotional motion coefficients and an art style source picture. Moreover, to better preserve image details and avoid artifacts, we provide StyleGAN with the multi-scale content features extracted from the identity image and refine its intermediate feature maps by the designed content encoder and refinement network, respectively. Extensive experimental results demonstrate our method outperforms existing state-of-the-art methods in terms of audio-lip synchronization and performance of both emotion style and art style.

FlowVQTalker: High-Quality Emotional Talking Face Generation through Normalizing Flow and Quantization

Generating emotional talking faces is a practical yet challenging endeavor. To create a lifelike avatar, we draw upon two critical insights from a human perspective: 1) The connection between audio and the non-deterministic facial dynamics, encompassing expressions, blinks, poses, should exhibit synchronous and one-to-many mapping. 2) Vibrant expressions are often accompanied by emotion-aware high-definition (HD) textures and finely detailed teeth. However, both aspects are frequently overlooked by existing methods. To this end, this paper proposes using normalizing Flow and Vector-Quantization modeling to produce emotional talking faces that satisfy both insights concurrently (FlowVQTalker). Specifically, we develop a flow-based coefficient generator that encodes the dynamics of facial emotion into a multi-emotion-class latent space represented as a mixture distribution. The generation process commences with random sampling from the modeled distribution, guided by the accompanying audio, enabling both lip-synchronization and the uncertain nonverbal facial cues generation. Furthermore, our designed vector-quantization image generator treats the creation of expressive facial images as a code query task, utilizing a learned codebook to provide rich, high-quality textures that enhance the emotional perception of the results. Extensive experiments are conducted to showcase the effectiveness of our approach.

Vosh: Voxel-Mesh Hybrid Representation for Real-Time View Synthesis

The neural radiance field (NeRF) has emerged as a prominent methodology for synthesizing realistic images of novel views. While neural radiance representations based on voxels or mesh individually offer distinct advantages, excelling in either rendering quality or speed, each has limitations in the other aspect. In response, we propose a pioneering hybrid representation named Vosh, seamlessly combining both voxel and mesh components in hybrid rendering for view synthesis. Vosh is meticulously crafted by optimizing the voxel grid of NeRF, strategically with selected voxels replaced by mesh. Therefore, it excels in fast rendering scenes with simple geometry and textures through its mesh component, while simultaneously enabling high-quality rendering in intricate regions by leveraging voxel component. The flexibility of Vosh is showcased through the ability to adjust hybrid ratios, providing users the ability to control the balance between rendering quality and speed based on flexible usage. Experimental results demonstrates that our method achieves commendable trade-off between rendering quality and speed, and notably has real-time performance on mobile devices.

Multi-Scale Implicit Transformer with Re-parameterize for  Arbitrary-Scale Super-Resolution

Recently, the methods based on implicit neural representations have shown excellent capabilities for arbitrary-scale super-resolution (ASSR). Although these methods represent the features of an image by generating latent codes, these latent codes are difficult to adapt for different magnification factors of super-resolution, which seriously affects their performance. Addressing this, we design Multi-Scale Implicit Transformer (MSIT), consisting of an Multi-scale Neural Operator (MSNO) and Multi-Scale Self-Attention (MSSA). Among them, MSNO obtains multi-scale latent codes through feature enhancement, multi-scale characteristics extraction, and multi-scale characteristics merging. MSSA further enhances the multi-scale characteristics of latent codes, resulting in better performance. Furthermore, to improve the performance of network, we propose the Re-Interaction Module (RIM) combined with the cumulative training strategy to improve the diversity of learned information for the network. We have systematically introduced multi-scale characteristics for the first time in ASSR, extensive experiments are performed to validate the effectiveness of MSIT, and our method achieves state-of-the-art performance in arbitrary super-resolution tasks.

Unraveling the Mystery of Scaling Laws: Part I

Scaling law principles indicate a power-law correlation between loss and variables such as model size, dataset size, and computational resources utilized during training. These principles play a vital role in optimizing various aspects of model pre-training, ultimately contributing to the success of large language models such as GPT-4, Llama and Gemini. However, the original scaling law paper by OpenAI did not disclose the complete details necessary to derive the precise scaling law formulas, and their conclusions are only based on models containing up to 1.5 billion parameters. Though some subsequent works attempt to unveil these details and scale to larger models, they often neglect the training dependency of important factors such as the learning rate, context length and batch size, leading to their failure to establish a reliable formula for predicting the test loss trajectory. In this technical report, we confirm that the scaling law formulations proposed in the original OpenAI paper remain valid when scaling the model size up to 33 billion, but the constant coefficients in these formulas vary significantly with the experiment setup. We meticulously identify influential factors and provide transparent, step-by-step instructions to estimate all constant terms in scaling-law formulas by training on models with only 1M~60M parameters. Using these estimated formulas, we showcase the capability to accurately predict various attributes for models with up to 33B parameters before their training, including (1) the minimum possible test loss; (2) the minimum required training steps and processed tokens to achieve a specific loss; (3) the critical batch size with an optimal time/computation trade-off at any loss value; and (4) the complete test loss trajectory with arbitrary batch size.

V3D: Video Diffusion Models are Effective 3D Generators

Automatic 3D generation has recently attracted widespread attention. Recent methods have greatly accelerated the generation speed, but usually produce less-detailed objects due to limited model capacity or 3D data. Motivated by recent advancements in video diffusion models, we introduce V3D, which leverages the world simulation capacity of pre-trained video diffusion models to facilitate 3D generation. To fully unleash the potential of video diffusion to perceive the 3D world, we further introduce geometrical consistency prior and extend the video diffusion model to a multi-view consistent 3D generator. Benefiting from this, the state-of-the-art video diffusion model could be fine-tuned to generate 360degree orbit frames surrounding an object given a single image. With our tailored reconstruction pipelines, we can generate high-quality meshes or 3D Gaussians within 3 minutes. Furthermore, our method can be extended to scene-level novel view synthesis, achieving precise control over the camera path with sparse input views. Extensive experiments demonstrate the superior performance of the proposed approach, especially in terms of generation quality and multi-view consistency. Our code is available at https://github.com/heheyas/V3D

Multistep Consistency Models

Diffusion models are relatively easy to train but require many steps to generate samples. Consistency models are far more difficult to train, but generate samples in a single step. In this paper we propose Multistep Consistency Models: A unification between Consistency Models (Song et al., 2023) and TRACT (Berthelot et al., 2023) that can interpolate between a consistency model and a diffusion model: a trade-off between sampling speed and sampling quality. Specifically, a 1-step consistency model is a conventional consistency model whereas we show that a \infty-step consistency model is a diffusion model. Multistep Consistency Models work really well in practice. By increasing the sample budget from a single step to 2-8 steps, we can train models more easily that generate higher quality samples, while retaining much of the sampling speed benefits. Notable results are 1.4 FID on Imagenet 64 in 8 step and 2.1 FID on Imagenet128 in 8 steps with consistency distillation. We also show that our method scales to a text-to-image diffusion model, generating samples that are very close to the quality of the original model. 

DEADiff: An Efficient Stylization Diffusion Model with Disentangled Representations

The diffusion-based text-to-image model harbors immense potential in transferring reference style. However, current encoder-based approaches significantly impair the text controllability of text-to-image models while transferring styles. In this paper, we introduce \textit{DEADiff} to address this issue using the following two strategies: 1) a mechanism to decouple the style and semantics of reference images. The decoupled feature representations are first extracted by Q-Formers which are instructed by different text descriptions. Then they are injected into mutually exclusive subsets of cross-attention layers for better disentanglement. 2) A non-reconstructive learning method. The Q-Formers are trained using paired images rather than the identical target, in which the reference image and the ground-truth image are with the same style or semantics. We show that DEADiff attains the best visual stylization results and optimal balance between the text controllability inherent in the text-to-image model and style similarity to the reference image, as demonstrated both quantitatively and qualitatively.

The pitfalls of next-token prediction

Can a mere next-token predictor faithfully model human intelligence? We crystallize this intuitive concern, which is fragmented in the literature. As a starting point, we argue that the two often-conflated phases of next-token prediction -- autoregressive inference and teacher-forced training -- must be treated distinctly. The popular criticism that errors can compound during autoregressive inference, crucially assumes that teacher-forcing has learned an accurate next-token predictor. This assumption sidesteps a more deep-rooted problem we expose: in certain classes of tasks, teacher-forcing can simply fail to learn an accurate next-token predictor in the first place. We describe a general mechanism of how teacher-forcing can fail, and design a minimal planning task where both the Transformer and the Mamba architecture empirically fail in that manner -- remarkably, despite the task being straightforward to learn. We provide preliminary evidence that this failure can be resolved when training to predict multiple tokens in advance. We hope this finding can ground future debates and inspire explorations beyond the next-token prediction paradigm.

BrushNet: A Plug-and-Play Image Inpainting Model with Decomposed Dual-Branch Diffusion

Image inpainting, the process of restoring corrupted images, has seen significant advancements with the advent of diffusion models (DMs). Despite these advancements, current DM adaptations for inpainting, which involve modifications to the sampling strategy or the development of inpainting-specific DMs, frequently suffer from semantic inconsistencies and reduced image quality. Addressing these challenges, our work introduces a novel paradigm: the division of masked image features and noisy latent into separate branches. This division dramatically diminishes the model's learning load, facilitating a nuanced incorporation of essential masked image information in a hierarchical fashion. Herein, we present BrushNet, a novel plug-and-play dual-branch model engineered to embed pixel-level masked image features into any pre-trained DM, guaranteeing coherent and enhanced image inpainting outcomes. Additionally, we introduce BrushData and BrushBench to facilitate segmentation-based inpainting training and performance assessment. Our extensive experimental analysis demonstrates BrushNet's superior performance over existing models across seven key metrics, including image quality, mask region preservation, and textual coherence.

VideoMamba: State Space Model for Efficient Video Understanding

Addressing the dual challenges of local redundancy and global dependencies in video understanding, this work innovatively adapts the Mamba to the video domain. The proposed VideoMamba overcomes the limitations of existing 3D convolution neural networks and video transformers. Its linear-complexity operator enables efficient long-term modeling, which is crucial for high-resolution long video understanding. Extensive evaluations reveal VideoMamba's four core abilities: (1) Scalability in the visual domain without extensive dataset pretraining, thanks to a novel self-distillation technique; (2) Sensitivity for recognizing short-term actions even with fine-grained motion differences; (3) Superiority in long-term video understanding, showcasing significant advancements over traditional feature-based models; and (4) Compatibility with other modalities, demonstrating robustness in multi-modal contexts. Through these distinct advantages, VideoMamba sets a new benchmark for video understanding, offering a scalable and efficient solution for comprehensive video understanding.