Compensating Spatiotemporally Inconsistent Observations for Online Dynamic 3D Gaussian Splatting

Online reconstruction of dynamic scenes is significant as it enables learning scenes from live-streaming video inputs, while existing offline dynamic reconstruction methods rely on recorded video inputs. However, previous online reconstruction approaches have primarily focused on efficiency and rendering quality, overlooking the temporal consistency of their results, which often contain noticeable artifacts in static regions. This paper identifies that errors such as noise in real-world recordings affect temporal inconsistency in online reconstruction. We propose a method that enhances temporal consistency in online reconstruction from observations with temporal inconsistency which is inevitable in cameras. We show that our method restores the ideal observation by subtracting the learned error. We demonstrate that applying our method to various baselines significantly enhances both temporal consistency and rendering quality across datasets. Code, video results, and checkpoints are available at https://bbangsik13.github.io/OR2.

4D Scaffold Gaussian Splatting for Memory Efficient Dynamic Scene Reconstruction

Existing 4D Gaussian methods for dynamic scene reconstruction offer high visual fidelity and fast rendering. However, these methods suffer from excessive memory and storage demands, which limits their practical deployment. This paper proposes a 4D anchor-based framework that retains visual quality and rendering speed of 4D Gaussians while significantly reducing storage costs. Our method extends 3D scaffolding to 4D space, and leverages sparse 4D grid-aligned anchors with compressed feature vectors. Each anchor models a set of neural 4D Gaussians, each of which represent a local spatiotemporal region. In addition, we introduce a temporal coverage-aware anchor growing strategy to effectively assign additional anchors to under-reconstructed dynamic regions. Our method adjusts the accumulated gradients based on Gaussians' temporal coverage, improving reconstruction quality in dynamic regions. To reduce the number of anchors, we further present enhanced formulations of neural 4D Gaussians. These include the neural velocity, and the temporal opacity derived from a generalized Gaussian distribution. Experimental results demonstrate that our method achieves state-of-the-art visual quality and 97.8% storage reduction over 4DGS.

HARIVO: Harnessing Text-to-Image Models for Video Generation

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

FLoD: Integrating Flexible Level of Detail into 3D Gaussian Splatting for Customizable Rendering

3D Gaussian Splatting (3DGS) achieves fast and high-quality renderings by using numerous small Gaussians, which leads to significant memory consumption. This reliance on a large number of Gaussians restricts the application of 3DGS-based models on low-cost devices due to memory limitations. However, simply reducing the number of Gaussians to accommodate devices with less memory capacity leads to inferior quality compared to the quality that can be achieved on high-end hardware. To address this lack of scalability, we propose integrating a Flexible Level of Detail (FLoD) to 3DGS, to allow a scene to be rendered at varying levels of detail according to hardware capabilities. While existing 3DGSs with LoD focus on detailed reconstruction, our method provides reconstructions using a small number of Gaussians for reduced memory requirements, and a larger number of Gaussians for greater detail. Experiments demonstrate our various rendering options with tradeoffs between rendering quality and memory usage, thereby allowing real-time rendering across different memory constraints. Furthermore, we show that our method generalizes to different 3DGS frameworks, indicating its potential for integration into future state-of-the-art developments. Project page: https://3dgs-flod.github.io/flod.github.io/

Rethinking Open-Vocabulary Segmentation of Radiance Fields in 3D Space

Understanding the 3D semantics of a scene is a fundamental problem for various scenarios such as embodied agents. While NeRFs and 3DGS excel at novel-view synthesis, previous methods for understanding their semantics have been limited to incomplete 3D understanding: their segmentation results are 2D masks and their supervision is anchored at 2D pixels. This paper revisits the problem set to pursue a better 3D understanding of a scene modeled by NeRFs and 3DGS as follows. 1) We directly supervise the 3D points to train the language embedding field. It achieves state-of-the-art accuracy without relying on multi-scale language embeddings. 2) We transfer the pre-trained language field to 3DGS, achieving the first real-time rendering speed without sacrificing training time or accuracy. 3) We introduce a 3D querying and evaluation protocol for assessing the reconstructed geometry and semantics together. Code, checkpoints, and annotations will be available online. Project page: https://hyunji12.github.io/Open3DRF

Eye-for-an-eye: Appearance Transfer with Semantic Correspondence in Diffusion Models

As pretrained text-to-image diffusion models have become a useful tool for image synthesis, people want to specify the results in various ways. In this paper, we introduce a method to produce results with the same structure of a target image but painted with colors from a reference image, i.e., appearance transfer, especially following the semantic correspondence between the result and the reference. E.g., the result wing takes color from the reference wing, not the reference head. Existing methods rely on the query-key similarity within self-attention layer, usually producing defective results. To this end, we propose to find semantic correspondences and explicitly rearrange the features according to the semantic correspondences. Extensive experiments show the superiority of our method in various aspects: preserving the structure of the target and reflecting the color from the reference according to the semantic correspondences, even when the two images are not aligned.

Per-Gaussian Embedding-Based Deformation for Deformable 3D Gaussian Splatting

As 3D Gaussian Splatting (3DGS) provides fast and high-quality novel view synthesis, it is a natural extension to deform a canonical 3DGS to multiple frames. However, previous works fail to accurately reconstruct dynamic scenes, especially 1) static parts moving along nearby dynamic parts, and 2) some dynamic areas are blurry. We attribute the failure to the wrong design of the deformation field, which is built as a coordinate-based function. This approach is problematic because 3DGS is a mixture of multiple fields centered at the Gaussians, not just a single coordinate-based framework. To resolve this problem, we define the deformation as a function of per-Gaussian embeddings and temporal embeddings. Moreover, we decompose deformations as coarse and fine deformations to model slow and fast movements, respectively. Also, we introduce an efficient training strategy for faster convergence and higher quality. Project page: https://jeongminb.github.io/e-d3dgs/

Semantic Image Synthesis with Unconditional Generator

Semantic image synthesis (SIS) aims to generate realistic images that match given semantic masks. Despite recent advances allowing high-quality results and precise spatial control, they require a massive semantic segmentation dataset for training the models. Instead, we propose to employ a pre-trained unconditional generator and rearrange its feature maps according to proxy masks. The proxy masks are prepared from the feature maps of random samples in the generator by simple clustering. The feature rearranger learns to rearrange original feature maps to match the shape of the proxy masks that are either from the original sample itself or from random samples. Then we introduce a semantic mapper that produces the proxy masks from various input conditions including semantic masks. Our method is versatile across various applications such as free-form spatial editing of real images, sketch-to-photo, and even scribble-to-photo. Experiments validate advantages of our method on a range of datasets: human faces, animal faces, and buildings.

Visual Style Prompting with Swapping Self-Attention

In the evolving domain of text-to-image generation, diffusion models have emerged as powerful tools in content creation. Despite their remarkable capability, existing models still face challenges in achieving controlled generation with a consistent style, requiring costly fine-tuning or often inadequately transferring the visual elements due to content leakage. To address these challenges, we propose a novel approach, \ours, to produce a diverse range of images while maintaining specific style elements and nuances. During the denoising process, we keep the query from original features while swapping the key and value with those from reference features in the late self-attention layers. This approach allows for the visual style prompting without any fine-tuning, ensuring that generated images maintain a faithful style. Through extensive evaluation across various styles and text prompts, our method demonstrates superiority over existing approaches, best reflecting the style of the references and ensuring that resulting images match the text prompts most accurately. Our project page is available https://curryjung.github.io/VisualStylePrompt/.

Attribute Based Interpretable Evaluation Metrics for Generative Models

When the training dataset comprises a 1:1 proportion of dogs to cats, a generative model that produces 1:1 dogs and cats better resembles the training species distribution than another model with 3:1 dogs and cats. Can we capture this phenomenon using existing metrics? Unfortunately, we cannot, because these metrics do not provide any interpretability beyond "diversity". In this context, we propose a new evaluation protocol that measures the divergence of a set of generated images from the training set regarding the distribution of attribute strengths as follows. Single-attribute Divergence (SaD) measures the divergence regarding PDFs of a single attribute. Paired-attribute Divergence (PaD) measures the divergence regarding joint PDFs of a pair of attributes. They provide which attributes the models struggle. For measuring the attribute strengths of an image, we propose Heterogeneous CLIPScore (HCS) which measures the cosine similarity between image and text vectors with heterogeneous initial points. With SaD and PaD, we reveal the following about existing generative models. ProjectedGAN generates implausible attribute relationships such as a baby with a beard even though it has competitive scores of existing metrics. Diffusion models struggle to capture diverse colors in the datasets. The larger sampling timesteps of latent diffusion model generate the more minor objects including earrings and necklaces. Stable Diffusion v1.5 better captures the attributes than v2.1. Our metrics lay a foundation for explainable evaluations of generative models.