Research & Education

[navtest] NAVSIM v1 End-to-End Driving is an AI simulation environment hosted on Hugging Face Spaces, designed for autonomous vehicle research and development. This platform allows users to participate in competitions, manage their submissions, and track their performance on leaderboards. It provides essential information regarding the dataset used for the simulations, competition rules, and details about individual submissions. The tool is specifically tailored for benchmarking end-to-end driving models, offering a standardized environment for researchers and developers to test and compare their AI algorithms. Its focus on competition and leaderboards makes it a valuable resource for advancing the field of autonomous driving.

Reward Bench Leaderboard is a platform hosted on Hugging Face Spaces by allenai, designed for ranking and comparing AI models using reward benchmarks. It provides a comprehensive leaderboard where users can browse different models, filter them by name using regex, and categorize them by type. The platform showcases model performance across various evaluation domains, offering insights into their capabilities. Additionally, users can view random example prompts and responses to better understand model behavior. This tool is invaluable for researchers and engineers who need to track and assess the performance of AI models in a standardized manner.

Splatt3R is the official implementation of a research project focused on zero-shot Gaussian Splatting from uncalibrated image pairs. This feed-forward model is designed to directly predict 3D Gaussians from standard images, eliminating the need for complex calibration processes. It is particularly useful for computer vision and 3D graphics applications where rapid 3D scene reconstruction from minimal input is critical. The tool provides an initial codebase, a research paper, a project webpage, and a Gradio demo for easy experimentation. Users can set up an Anaconda environment, compile CUDA kernels, and utilize pretrained models and data from ScanNet++ to train their own models or generate 3D scene representations.

MotionModel is an AI tool hosted on Hugging Face that specializes in analyzing motion within video content. It provides detailed visualizations of motion flow, neural activation, and attention, offering insights into how movement is perceived and processed. Users can upload their own videos to the platform and utilize adjustable sliders to refine their focus on particular areas of interest within the footage. This capability makes it a valuable resource for researchers and developers working with video analysis and computer vision, allowing for in-depth exploration and testing of self-attention-based motion models.

FarmingGame is an AI simulation tool hosted on Hugging Face Spaces, designed for creating and experimenting with farming game simulations. While the specific features of the FarmingGame application itself are not detailed on the provided pricing page, the platform it resides on, Hugging Face, offers extensive resources for AI development and deployment. Users can leverage Hugging Face's infrastructure for model hosting, dataset management, and running AI applications. The platform provides various hardware options for Spaces, including CPU and GPU instances, and offers dedicated Inference Endpoints for deploying models in production. This makes FarmingGame a potential sandbox for AI enthusiasts and game developers to explore and test AI concepts within a gaming environment, utilizing the robust computational backend of Hugging Face.

4d-gaussian-splatting is an open-source implementation for real-time photorealistic dynamic scene representation and rendering, based on the ICLR 2024 paper "Real-time Photorealistic Dynamic Scene Representation and Rendering with 4D Gaussian Splatting." This tool allows users to model dynamic scenes using native 4D Gaussian primitives, offering a coherent integrated approach to space and time dimensions. It builds upon the principles of 3D Gaussian Splatting and provides a dedicated rendering pipeline. The project includes resources for data preparation using datasets like DyNeRF and DNeRF, and offers scripts for training models. It's ideal for researchers and developers working on advanced 3D and animation projects.

SparseDrive introduces a sparse-centric paradigm for end-to-end autonomous driving, focusing on sparse scene representation to unify various tasks. It features a symmetric sparse perception model that integrates detection, tracking, and online mapping. The tool also includes a parallel motion planner designed for both motion prediction and planning, incorporating a hierarchical planning selection strategy with a collision-aware rescore module to enhance safety. SparseDrive demonstrates superior performance on the nuScenes benchmark, outperforming previous state-of-the-art methods in all metrics, particularly collision rate, while maintaining high training and inference efficiency. It is an open-source project, making its code and models accessible for research and development.

Segformer B0 Segments Sidewalk Finetuned is an AI tool designed for detailed image segmentation, specifically trained to identify and highlight elements like roads, sidewalks, people, and vehicles. Users can upload an image, and the application processes it to provide a visual overlay of these segmented objects. This capability is particularly useful for urban environment analysis, contributing to applications in autonomous vehicle development and pedestrian safety initiatives through accurate sidewalk segmentation. The tool offers a straightforward way to visualize and understand the composition of urban scenes.

SoloAudio is an innovative AI tool developed by OpenSound, available as a Hugging Face Space, designed to intelligently separate specific sounds from complex audio mixtures. Users can upload an audio file and then provide a text prompt describing the desired sound they wish to isolate. The application processes the input and generates a new audio file containing only the specified sound, effectively removing other elements from the original recording. This capability is highly beneficial for audio editing, sound design, and various research applications in audio processing, offering a streamlined approach to sound extraction.

SoloSpeech is an advanced AI tool designed for target speech extraction, enabling users to isolate and extract specific voices from audio recordings. By uploading an audio file containing multiple voices and a short sample of the desired speaker, the application processes the input to return a clean audio file with only the target speech. This state-of-the-art tool is particularly useful for tasks requiring precise voice isolation, such as enhancing audio quality, conducting speech processing research, or developing applications that rely on clean, isolated speech. Its intuitive interface on Hugging Face Spaces makes it accessible for various users looking to refine audio content.

Small Object Detection with YOLO11 is an AI tool hosted on Hugging Face Spaces, designed for identifying small objects within images. It leverages the YOLO (You Only Look Once) architecture, specifically YOLO11, in conjunction with SAHI (Slicing Aided Hyper Inference) to enhance detection capabilities. Users can upload their own images or utilize provided examples to test the tool. Key features include the ability to adjust confidence thresholds and slice sizes, which are crucial for optimizing detection accuracy and ensuring comprehensive coverage of small objects in various scenarios. This tool is suitable for researchers, developers, and anyone interested in advanced object detection techniques.

Small Object Detection with YOLO26 is an AI tool hosted on Hugging Face Spaces, designed for advanced object detection and segmentation tasks. It leverages the power of YOLO26 and SAHI (Slicing Aided Hyper Inference) to accurately identify and segment small objects within images. Users can upload an image, select a preferred YOLO26 detection or segmentation model, and the application will perform both standard and SAHI-sliced inference. The results are returned as two versions of the original image, clearly marked with bounding boxes and segmentation masks, making it ideal for research, development, and educational exploration of computer vision techniques.

Small Object Detection with YOLOX is an AI tool hosted on Hugging Face Spaces, designed for identifying small objects within images. It leverages the YOLOX architecture and offers an enhanced SAHI+YOLOX method for improved detection capabilities. Users can upload or select an image, set parameters like slice size and overlap ratio, and then perform predictions to compare the results between standard YOLOX and SAHI+YOLOX. This tool is valuable for researchers, developers, and educators interested in experimenting with advanced object detection techniques and understanding the benefits of SAHI integration for small object detection.

webdemo-fridge-detection is an AI tool designed for object detection, specifically within the context of a refrigerator. Hosted on Hugging Face Spaces by dnth, the tool's intended purpose is to analyze images and identify items inside a fridge. However, based on the live website content, the application is currently experiencing a runtime error, indicating a module not found issue. This prevents users from interacting with the tool and utilizing its object detection capabilities. While the concept suggests utility for research, educational demonstrations, or testing object detection models, its current operational status is non-functional.

WebGPU Video Object Detection is an AI tool hosted on Hugging Face Spaces that leverages your webcam to perform real-time object detection. This application displays the detection results directly on a canvas, providing immediate visual feedback. Users have the flexibility to fine-tune various parameters, including the stream scale, image size, and detection threshold, to achieve optimal performance and accuracy for their specific needs. This makes it a versatile tool for experimenting with real-time object detection, potentially useful for developers and researchers working with computer vision models and WebGPU technology. It offers a hands-on way to interact with and understand the capabilities of object detection in a live video feed.

VLM R1 OVD is an AI tool designed for open-vocabulary object detection, hosted as a Hugging Face Space. Users can upload an image and provide a list of objects they wish to detect within that image. The application then processes the input, identifies the specified objects, and draws bounding boxes around them. Additionally, it provides a 'thinking process' and an answer, offering insights into how the detection was performed. This tool leverages the VLM-R1 model for its object detection capabilities, making it suitable for tasks requiring flexible and dynamic object identification without being limited to pre-defined categories.

Webrtc Yolov10N is a computer vision tool designed for real-time object detection, leveraging the YOLOv10 model. Hosted as a Hugging Face Space, it enables users to stream video directly from their webcam and observe objects being detected in real-time. A key feature is the ability to adjust the confidence threshold, giving users control over the sensitivity of the object detection process. This makes it suitable for various computer vision projects where immediate visual feedback and customizable detection parameters are crucial. The tool is implemented within a Gradio interface, providing an accessible platform for interaction.

YOLO ARENA is a powerful tool hosted on Hugging Face designed for comparing the performance of leading object detection models. Users can upload any image and fine-tune detection strictness by adjusting confidence and Intersection over Union (IoU) sliders. The application runs five pre-trained YOLO models (v8, v9, v10, v11, and RF-DETR) on the uploaded image, providing a direct comparison of their detection capabilities. This allows developers and researchers to evaluate and benchmark different object detection algorithms efficiently, making it an invaluable resource for understanding model strengths and weaknesses in various scenarios.

Zero Shot Object Detection Arena is an AI tool hosted on Hugging Face Spaces that enables users to perform object detection on images. Users can upload an image and provide object prompts to identify and label specific objects within it. The platform then processes the image using four different object detection models, providing annotated images with bounding boxes and labels, along with the inference times for each model. This allows for quick comparison and evaluation of various zero-shot object detection capabilities without the need for extensive training data.

Zero Shot Video Classification is an AI tool hosted on Hugging Face Spaces that enables users to classify videos into various categories without the need for pre-trained models on those specific categories. This tool leverages zero-shot learning techniques, allowing for flexible and dynamic video content analysis. Users can input a YouTube URL or a local video file, and the system attempts to classify the video based on provided candidate labels. While the live application currently shows a runtime error, its intended functionality is to provide a quick and accessible way to perform video classification for various applications, from content moderation to data analysis.

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.

Experimental Moondream WebGPU offers an innovative platform for rendering 3D graphics within a web browser, leveraging the power of WebGPU technology. This tool is designed for users who need to display 3D models and textures with high performance directly on the web. It provides a robust environment for experimenting with advanced graphics rendering techniques and exploring the capabilities of WebGPU. Ideal for developers and researchers, it facilitates the testing and visualization of complex 3D data without requiring specialized desktop applications. The application is hosted on Hugging Face Spaces, making it easily accessible and shareable for collaborative projects and demonstrations.

Neural Acoustic Distance is an AI tool available as a Hugging Face Space, designed for analyzing and comparing audio data, specifically single-word WAV files. Users can upload two audio files and select a wav2vec 2.0 model layer to compute the neural acoustic distance between them. The tool then provides a frame-by-frame plot, illustrating how the pronunciations differ. This functionality is particularly useful for researchers and developers in audio engineering, phonetics, or speech technology who need to quantitatively assess and visualize subtle acoustic variations between spoken words. It offers a practical way to gain insights into speech patterns and model performance.

The NAVSIM v2 End-to-End Driving Challenge 2025 is an AI simulation tool designed for advanced research in autonomous vehicle technology. It offers a comprehensive simulated driving environment, crucial for testing and training AI driver models. The platform serves as a hub for competition participants, providing detailed information on rules, datasets, and a real-time leaderboard. Users can manage their submissions, track their progress, and update team details, fostering a dynamic and competitive research environment. This tool is particularly valuable for robotics researchers and developers focused on pushing the boundaries of autonomous driving AI.