Content & Design

SHARP - 3D Gaussian Scene Prediction from Apple is an AI tool available as a Hugging Face Space that transforms static 2D images into dynamic 3D Gaussian Splat scenes. This application allows users to upload any 2D picture and generate a 3D scene from it, offering control over various output parameters. Users can select desired camera movement, output resolution, the number of frames, and frames per second (FPS). Additionally, the tool provides the option to render a video preview of the generated 3D scene, simplifying the creation of immersive 3D environments from simple images.

E2E FT Marigold for Normals is an AI tool hosted on Hugging Face that specializes in generating surface normals from uploaded images. Users can input an image and receive two outputs: the raw data of the surface normals and a corresponding colored map. This tool is particularly useful for tasks requiring detailed surface information, such as 3D reconstruction, computer vision research, or graphics applications. It is licensed under Apache-2.0, making it accessible for various projects. The platform leverages Hugging Face's infrastructure, which offers different pricing tiers for storage, compute, and inference, catering to both individual developers and enterprise teams.

pytorch-openpose offers a PyTorch implementation of the popular OpenPose framework, enabling robust body and hand pose estimation. This tool is particularly useful for researchers and developers working with computer vision and human-computer interaction, as it facilitates the conversion of existing OpenPose Caffe models to PyTorch. Beyond body and hand pose, it also supports the implementation of face keypoint detection, following similar procedures used for hand detection. The project provides clear instructions for setting up a Python 3.7 environment, installing necessary dependencies, and downloading pre-trained models. Demos are available for webcam, image, and video processing, making it accessible for immediate experimentation and integration into various projects.

PoseEstimationForMobile is an open-source project designed for real-time single-person pose estimation on Android and iOS devices. It leverages CPM and Hourglass models, implemented with TensorFlow, and incorporates inverted residuals (MobileNet V2) for optimized, real-time inference. The repository includes code for training both CPM and Hourglass models, along with demo source code for Android and iOS. This allows developers to integrate pose estimation capabilities into their mobile applications with high performance. The project provides pre-trained models and detailed instructions for setting up training environments, converting models for mobile deployment (Mace, TFLite, CoreML), and benchmarking performance across various mobile chipsets.

PVN3D is the official source code for "PVN3D: A Deep Point-wise 3D Keypoints Hough Voting Network for 6DoF Pose Estimation," a research paper presented at CVPR 2020. This open-source project enables researchers and developers to implement and experiment with advanced 6DoF pose estimation techniques using 3D keypoints. It supports training and evaluation on popular datasets like LineMOD and YCB-Video, and includes pre-trained models for various objects. The tool also offers guidance for adapting the framework to new datasets, making it a valuable resource for academic research and development in computer vision and robotics. It is built with Python and PyTorch, requiring specific CUDA and Python environment setups.

Marigold Depth Estimation is an AI tool hosted on Hugging Face Spaces that allows users to upload a single image and generate a visual depth map. The application also provides a 16-bit depth file for download, enabling further processing or integration into other projects. Users can fine-tune the depth estimation process by adjusting various settings, including the ensemble size and denoising steps, to achieve optimal quality and detail in the generated depth maps. This tool is particularly useful for applications requiring 3D scene understanding, computer vision, and graphics processing, offering a straightforward way to extract depth information from 2D images.

Marigold Normals Estimation is an AI tool hosted on Hugging Face Spaces, developed by the Photogrammetry and Remote Sensing Lab of ETH Zurich. It allows users to upload images and compute surface normals in real time. This process helps in understanding the 3D orientation of surfaces within a 2D image, which is crucial for various computer vision and graphics applications. The tool provides adjustable settings such as ensemble size and denoising steps, enabling users to fine-tune the estimation process and achieve more refined results. It is designed for real-time processing, making it efficient for quick analyses and iterative adjustments.

Marigold-LCM Depth Estimation (Deprecated) was an AI tool designed to generate detailed depth maps from single images. Users could upload a picture and receive a visualization indicating the distance of objects from the camera. This system was known for providing fast and accurate depth estimations, which could be valuable for various applications requiring 3D scene understanding. While the tool is now deprecated, its functionality focused on making complex depth estimation accessible to users through a straightforward interface.

Reachy Dance Duo is an innovative AI tool available as a Hugging Face Space that allows users to experience two Reachy Mini robots dancing in synchronization with music. This web application provides an engaging visual and auditory experience, letting you start or pause the demo audio at your convenience. Beyond simple playback, the tool offers on-screen controls to customize the visual appearance of the logo and precisely adjust the robots' positions, offering a degree of interactivity. It transforms any song into a dynamic robot dance party, making it a unique and entertaining demonstration of robotics and AI in action. The tool is designed for ease of use, providing a straightforward interface for immediate enjoyment.

Repainting 3D Assets is a specialized tool designed for modifying the textures and appearances of existing 3D models. This application enables users to customize their 3D assets, providing flexibility in design and aesthetic adjustments. While the live website content indicates the tool is currently sleeping due to inactivity, its core function is to facilitate the 'repainting' process, which implies changing colors, materials, or surface properties of 3D objects. This capability is particularly valuable for professionals in fields such as game development, architectural visualization, and 3D art, where unique and customized visual elements are crucial for projects.

Simple Animation Doodle is an interactive AI tool hosted on Hugging Face, designed for creating and animating basic drawings. Users can draw directly on a canvas and then apply simple animation features to their creations, making them come to life. This tool is ideal for individuals looking to experiment with animation, create quick visual concepts, or add a dynamic touch to their doodles without needing complex software. It's accessible and user-friendly, making it suitable for a wide range of creative applications from educational purposes to personal projects.

Stable Point-Aware 3D is an AI tool hosted on Hugging Face that enables users to generate 3D models from uploaded images. The platform allows for post-generation editing of the point cloud, providing flexibility in refining the 3D output. Once satisfied, users can download their final 3D models in multiple formats, making it suitable for various applications. This tool is designed for experimenting with point-aware 3D model generation techniques and exploring their capabilities and potential uses in research, education, and 3D content creation.

Versy AI is a Spatial Communication Platform designed to convert spatial ideas into immersive and shareable experiences. This tool empowers designers, studios, and teams to effectively present, iterate on, and secure approvals for their projects by allowing stakeholders to experience spaces virtually before they are physically built. By enabling users to "feel space before it exists," Versy AI streamlines the design and approval process, reducing the need for costly physical prototypes or extensive revisions. It focuses on creating engaging, interactive virtual environments that enhance communication and understanding of spatial concepts.

openMVG (open Multiple View Geometry) is a comprehensive open-source C++ library designed for 3D computer vision and Structure from Motion (SfM). It offers an end-to-end framework for 3D reconstruction from images, including libraries, binaries, and pipelines. The libraries provide access to features such as image manipulation, feature description and matching, camera models, robust estimation, and various SfM algorithms. Binaries handle unit tasks like scene initialization and reconstruction export, while pipelines chain these binaries for tasks like image matching and solving the SfM problem. OpenMVG is developed with a focus on reproducibility, maintainability, and ease of use, running on Android, iOS, Linux, macOS, and Windows.

The WebGL Gaussian Splat Viewer is an interactive application designed for visualizing 3D Gaussian splats directly within a web browser using WebGL technology. Users can easily control the camera through mouse, arrow keys, or touch gestures, enabling seamless navigation and exploration of complex 3D environments. This tool is particularly useful for individuals working with 3D graphics, researchers, and developers who need to inspect and interact with Gaussian splat models. Its web-based nature makes it accessible without requiring specialized software installations, offering a convenient way to share and review 3D content.

Avumi is currently in its pre-launch phase, as indicated by its 'Launching Soon' homepage. The website provides a contact form for users to sign up for email updates, promotions, and more, suggesting future plans for a product or service. The site also includes standard legal disclaimers regarding reCAPTCHA, Google's Privacy Policy, and Terms of Service, along with a copyright notice for 2025. Cookie usage is also mentioned for website traffic analysis and optimization. Beyond these basic details, specific features, pricing, or the exact nature of the AI tool are not yet disclosed.

Krea Realtime Video is an AI-powered tool designed for generating videos in real-time. Hosted on Hugging Face, it facilitates the creation of dynamic visual content, making it suitable for applications such as interactive art installations and various creative video projects. The tool emphasizes immediate feedback and generation, allowing users to experiment and produce video content on the fly.

LD T3D is an AI-powered tool that facilitates the creation of 3D models. It leverages artificial intelligence to generate 3D assets, streamlining the modeling process for its users. The tool is accessible for free on the Hugging Face platform, making it a valuable resource for individuals and professionals looking to integrate AI into their 3D design workflows. It caters to a diverse audience, including 3D artists, game developers, and AI researchers who require efficient 3D asset generation capabilities.

fast-gaussian-rasterization is a cutting-edge 3D Gaussian Splatting rasterizer engineered for high-performance rendering. Leveraging a geometry-shader-based approach and global CUDA sorting, this tool significantly accelerates rendering, offering a 5-10x speedup over traditional software CUDA rasterizers. It's particularly effective for scenarios with a high pixel-to-point ratio, such as large Gaussians and high-resolution rendering. The tool is designed for seamless integration, allowing users to replace existing `diff_gaussian_rasterization` imports with `fast_gauss` for immediate performance gains. It supports NVIDIA GPUs with CUDA and OpenGL interop, making it suitable for both Linux and Windows environments.

nerfplusplus is an open-source codebase designed to enhance Neural Radiance Fields (NeRF) for capturing and rendering large-scale, unbounded 360-degree scenes. It offers significant improvements over traditional NeRF methods, particularly for complex environments. The codebase supports multi-GPU training and inference through PyTorch DistributedDataParallel, enabling efficient processing of demanding tasks. An experimental feature for optimizing per-image autoexposure is also included. It provides tools for data preparation, including generating camera parameters with COLMAP SfM, scene normalization, and visualizing cameras in 3D to ensure compatibility and correctness.

OpenSplat is a free and open-source C++ implementation of 3D Gaussian splatting, designed for portability, efficiency, and speed. It can run on Windows, Mac, and Linux, with support for NVIDIA, AMD, and Apple (Metal) GPUs, as well as CPU-only operation (though significantly slower). The tool takes camera poses and sparse points from formats like COLMAP, OpenSfM, ODM, or nerfstudio projects to compute scene files (.ply or .splat). These generated files can then be imported into other software for viewing, editing, and rendering. OpenSplat is licensed under AGPLv3, allowing and encouraging commercial use.

DirectVoxGO is an open-source tool designed for fast radiance field reconstruction, leveraging direct voxel grid optimization. It significantly speeds up NeRF (Neural Radiance Fields) by replacing traditional MLPs with a voxel grid for volume densities and a dense feature grid with a shallow MLP for view-dependent colors. The tool includes a PyTorch CUDA extension for additional 2-3x speedup and an O(N) realization for the distortion loss, improving both training time and quality. It supports various datasets including bounded and unbounded inward-facing scenes, as well as forward-facing scenes, making it versatile for researchers and engineers in computer vision.

DeepMimic is an open-source framework that utilizes deep reinforcement learning to train simulated characters, such as humanoids and dogs, to imitate a variety of motion skills from mocap data. The tool provides a robust environment for researchers and developers to experiment with physics-based character control. While the codebase itself is deprecated, it served as the foundation for significant advancements in motion imitation, with its successor, MimicKit, offering faster and easier implementations. DeepMimic supports training policies for complex actions like spinkicks and locomotion, and allows for visualization of trained policies and mocap data playback. It requires a technical setup with dependencies like Bullet, Eigen, OpenGL, and Tensorflow, making it suitable for users with a strong programming background.

HyperPose is a powerful library designed for building high-performance custom human pose estimation applications. It stands out with its real-time capabilities, achieved through a sophisticated pose estimation engine that incorporates numerous system optimizations. These include pipeline parallelism, model inference with TensorRT, and CPU/GPU hybrid scheduling, leading to significantly higher FPS compared to other popular tools like OpenPose, TF-Pose, and OpenPifPaf. Beyond performance, HyperPose offers flexibility for developers, providing high-level Python APIs to customize training, evaluation, visualization, pre-processing, and post-processing. Users can also tailor model architectures and training datasets, and accelerate training with multiple GPUs, making it a versatile solution for advanced computer vision projects.