Data & Analytics

DINOv3 Web is an innovative tool designed for visualizing rich, dense image features directly within your web browser. Users can upload any picture, and the application extracts visual features without requiring server-side processing. As you interact with the image by moving your mouse or finger, an overlay or heatmap dynamically highlights patches that exhibit the most similarity to the selected area. This interactive visualization helps in understanding the underlying representations generated by the DINOv3 model, making it a valuable resource for researchers, data scientists, and developers working with computer vision models. The tool is hosted on Hugging Face Spaces and is licensed under Apache 2.0, promoting open access and collaboration.

Depth Pro is an AI tool designed for monocular metric depth estimation, allowing users to generate inverse depth maps from single images. This application highlights distances within a scene and provides the focal length in pixels, offering valuable insights into image composition and spatial relationships. Based on research, Depth Pro is particularly useful for real-time depth processing applications where quick and accurate depth information is crucial. It is available as a Hugging Face Space, making it accessible for users interested in computer vision and image analysis tasks. The tool aims to provide sharp depth maps efficiently.

DepthAnything AC is an AI tool designed for estimating depth from images and videos, providing a detailed 3D structural understanding of a scene. Users can upload their media files, and the application will process them to generate a corresponding depth map. A key feature is the ability to choose from different color maps, allowing for diverse visualizations of the depth information. This tool is based on the 'Depth Anything at Any Condition' paper, offering robust depth estimation capabilities. It is available as a Hugging Face Space, making it accessible for various applications requiring 3D scene understanding.

gradio_image_annotation V0.5.0 is a Gradio component designed for image annotation, enabling users to easily upload or capture images and then draw bounding boxes around objects of interest. This tool facilitates the labeling of these annotated areas, providing the output in a structured JSON format, which includes both the coordinates and the assigned labels. It is particularly useful for tasks requiring the creation of datasets for computer vision projects, such as object detection or image segmentation. The component simplifies the process of generating labeled data, making it accessible for developers and data scientists working on AI models.

BlenderProc is a powerful open-source tool designed to create photorealistic synthetic training images using a procedural Blender pipeline. It's ideal for generating large datasets for computer vision models, offering extensive features for loading diverse object formats like .obj, .ply, .blend, and BOP datasets. Users can procedurally set object poses, apply physics for collision checking, and manipulate materials and lighting. The tool supports rendering various image types including RGB, stereo, depth, normal, and segmentation images, and can write results to .hdf5 containers with COCO & BOP annotations. It provides comprehensive documentation, tutorials, and examples to help users get started with synthetic data generation.

Metavido, formerly known as Bibcam, is an innovative video subformat that allows for the direct embedding of camera metadata into video frames. It utilizes a burnt-in-barcode technique to achieve this, alongside integrating non-color planes such as depth information and human stencil through a squeezing method. This unique approach enables the recording, editing, and playback of AR-ready video clips without the common issue of desynchronization with external tracking data. The tool requires Unity 6 and a LiDAR-enabled iOS device for recording, making it suitable for developers and content creators working with augmented reality video. Users can capture Metavido clips via an encoder scene and play them back using a decoder scene, with options to adjust settings like frame rate.

caffe-yolo offers a Caffe implementation of the YOLO (You Only Look Once) real-time object detection system. This tool specifically supports YOLO v1 and includes batch normalization layers. The Caffe models used are not trained within Caffe but are converted from Darknet's original .weight files, ensuring compatibility and leveraging existing pre-trained models. The conversion process involves creating .prototxt files from Darknet's .cfg files, initializing the Caffe network, reading weights from Darknet, and then replacing initialized weights with the pre-trained ones. It provides scripts for creating .prototxt and .caffemodel files, and a main script for performing object detection on images. This makes it a valuable resource for developers and researchers working with object detection in a Caffe environment.

Number Recognizer is an AI tool hosted on Hugging Face that specializes in recognizing digits from images of house or door plates. Users can easily upload a picture containing a house or door number, select a preferred model checkpoint, and the application will quickly process the image to read the displayed digits. The tool then returns the recognized number as plain text, along with a status indicating the recognition outcome. This application is useful for tasks requiring automated number extraction from real-world images, offering a straightforward solution for digit recognition.

ner-annotator is a specialized Named Entity Recognition (NER) annotation tool designed to create training data for custom NER models with SpaCy. It provides an intuitive user interface for labelling entities in text, supporting both word-level and character-level annotation. Users can define custom labels with color-coding for enhanced clarity. The tool generates training data in a generic JSON format, making it readily usable for various tagging formats like IO, IOB, or IOBES. While no longer actively maintained, the web application and desktop versions (Linux and Windows) remain fully functional, offering features like keyboard shortcuts and the ability to import existing annotations for review. It also includes light and dark themes for user preference.

pytorch-pose is an open-source PyTorch toolkit designed for 2D single human pose estimation. It offers a comprehensive pipeline for training, inference, and evaluation, making it a valuable resource for researchers and developers in computer vision. The toolkit includes a robust dataloader with various data augmentation options, compatible with popular human pose databases such as MPII, LSP, and FLIC. Key features include multi-thread data loading, multi-GPU training support, a logger for tracking progress, and visualization of training and testing results. It is compatible with PyTorch 0.4.1/1.0 and provides detailed instructions for installation, data preparation, and usage, including testing with pre-trained models and evaluating PCKh@0.5 scores.

YOLOv11-RGBT offers a comprehensive single-stage multispectral object detection framework, extending the capabilities of YOLO models (from YOLOv3 to YOLOv13) and RTDETR to handle RGBT (Red, Green, Blue, Thermal) data. This project simplifies the configuration of visible and infrared datasets for multimodal object detection tasks, providing three distinct configuration methods. It supports multi-spectral object detection, keypoint detection, and instance segmentation. The framework is adaptable to various pixel-aligned images, including depth maps and SAR images, not just multispectral. Key features include support for TIFF images, 16-bit multi-spectral datasets with arbitrary channels, and various image formats like Gray, BGR, RGBT, and Multispectral with flexible channel configurations.

SAM3 VLM-FO1 is an AI tool designed for complex text label detection and object identification within images. Users can upload an image and provide natural language descriptions of the objects they wish to identify. The tool, leveraging SAM3 with VLM-FO1, then processes this input to highlight and label the specified objects directly on the image. This functionality makes it particularly useful for computer vision tasks and AI research, offering a practical application for detailed image annotation and understanding based on textual queries. It simplifies the process of identifying and categorizing visual elements through intuitive natural language interaction.

Grounding Dino Inference is an AI tool hosted on Hugging Face Spaces, designed for advanced object detection and image analysis. Users can upload an image and then provide text descriptions of the objects they wish to identify. The application leverages the Grounding Dino model to accurately locate and highlight these specified objects within the uploaded image. This tool is particularly useful for researchers and developers working in computer vision, offering a straightforward interface to perform complex inference tasks. It provides a practical demonstration of the Grounding Dino model's capabilities in identifying diverse objects based on natural language input.

entity-recognition-datasets is a valuable resource for researchers and developers working on named entity recognition (NER) and entity recognition tasks. This repository compiles a diverse collection of annotated datasets, spanning multiple languages, domains, and entity types. It serves as a crucial foundation for training and evaluating NER models, offering a wide array of corpora from news articles and social media to medical records and legal documents. The collection includes both readily available datasets and information on how to obtain those with licensing restrictions, often accompanied by conversion code to standard formats like CoNLL 2003. This makes it an essential tool for anyone looking to build or improve their NER systems across various applications and linguistic contexts.

ccv is a C-based/Cached/Core Computer Vision Library designed with a minimalism inspiration, making it easy to deploy and integrate into server-side environments. It is highly portable and embeddable, running on various platforms including Mac OSX, Linux, FreeBSD, Windows, iPhone, iPad, Android, and Raspberry Pi. The library implements a range of state-of-the-art algorithms, such as an image classifier, frontal face detector, object detectors for pedestrians and cars, text detection, and general object tracking. A key differentiator is its built-in cache mechanism for image preprocessing, which maintains a clean function interface while transparently handling redundant operations. ccv aims to provide high-performance, modern computer vision implementations, bridging the gap between older, battle-tested algorithms and newer, often MATLAB-based approaches.

Data-Labeling is an open-source tool designed for efficient processing and annotation of text data. It streamlines the text annotation process through simplified workflows and dynamic algorithm feedback, enabling users to quickly label keywords. The tool significantly reduces manual annotation costs and time by leveraging algorithms. Its methodology involves initial manual annotation to build a foundation, followed by automated annotation that feeds back into the manual process, and finally, manual correction to enhance accuracy and efficiency. Data-Labeling also features efficient annotation methods using various identifiers, shortcuts, and classification techniques, along with global algorithm calibration to reduce redundant work in multi-group annotation scenarios. It provides industry-specific vocabularies and supports various functionalities like article addition, filtering, export of segmented words, and detailed annotation logs.

RLHF-Reward-Modeling is an open-source repository offering comprehensive recipes and code for training reward models essential for Reinforcement Learning from Human Feedback (RLHF). The project supports various advanced techniques, including the classic Bradley-Terry reward model, pairwise preference models, and more recent innovations like Semi-Supervised Reward Modeling (SSRM) and ArmoRM for multi-objective reward modeling. It also provides code for process-supervised and outcome-supervised reward models, as well as decision-tree reward models. The repository emphasizes reproducibility, offering data, code, and hyperparameters for robust model training. It is designed to facilitate the development of state-of-the-art reward models, as evidenced by its models achieving top ranks on RewardBench.

SARDet_100K is a comprehensive dataset specifically designed for advancing research and development in synthetic aperture radar (SAR) object detection. This large-scale dataset facilitates the training and evaluation of models for multi-class rotated object detection tasks, a critical capability in various applications. Accepted at NeurIPS 2024 as a spotlight, SARDet_100K offers a robust foundation for researchers and developers working on complex SAR data analysis. Its focus on rotated object detection addresses a common challenge in SAR imagery, where objects can appear at various orientations, making it a valuable resource for developing more accurate and resilient detection algorithms.

DatologyAI is an advanced Data & Analytics platform designed to automatically curate and optimize training data for AI models. Leveraging cutting-edge research, it helps organizations train high-performing models more efficiently, reducing both time and computational costs. The platform addresses common issues like low-quality training data and the impossibility of manual data review at petabyte scale by automatically identifying and prioritizing the most valuable data points. This leads to faster model training, improved performance, and the ability to deploy smaller, more cost-effective models in production. DatologyAI offers data curation as a service, aiming to improve model performance, reduce deployment costs, and increase overall speed.

Chest x-ray HybridGNet Segmentation is a specialized tool designed for medical image analysis, specifically focusing on chest X-rays. It utilizes the HybridGNet model, incorporating image-to-graph skip connections to accurately segment and highlight key anatomical structures such as the lungs and heart. Users can upload a chest X-ray image, and the application will process it to provide detailed masks and landmarks for these organs. This tool demonstrates a training procedure derived from published research, making it a valuable resource for researchers, medical professionals, and data scientists interested in advanced medical imaging segmentation.

Chinese-CLIP Zero-Shot Image Classification is an AI tool designed for classifying images using Chinese language labels. It operates on a zero-shot learning paradigm, meaning it can classify images without requiring explicit training examples for each category. Users can upload an image and provide a list of potential labels in Chinese. The application then processes the input and returns the likelihood of each provided label being associated with the uploaded image. This tool is particularly useful for applications requiring image classification in a Chinese linguistic context, leveraging the capabilities of the Chinese-CLIP model. It is available as a demo on Hugging Face and is licensed under the MIT license.

Datasets Tagging is a Hugging Face Space application designed to streamline the process of creating and validating structured tags for datasets within the Hugging Face library. Users can input various details such as the dataset name, associated tasks, supported languages, creators, license information, and size. This functionality enables the generation of comprehensive and up-to-date metadata files, significantly improving dataset organization and documentation. The tool is particularly useful for maintaining consistency and discoverability across a large collection of datasets, making it an essential resource for data scientists and developers working with the Hugging Face ecosystem.

D-Fine is a state-of-the-art real-time object detector available as a Hugging Face Space. This application provides users with the capability to upload both images and videos, then apply object detection to identify elements within them. Users can fine-tune the detection process by specifying their preferred model checkpoint and adjusting the confidence threshold, offering a degree of customization over the results. For video content, the tool extends its detection capabilities to provide continuous analysis. Developed by the University of Science and Technology of China, D-Fine is a robust solution for anyone needing efficient and customizable object detection.

GLIP BLIP Ensemble Object Detection and VQA is a powerful tool that integrates Microsoft's GLIP and Salesforce's BLIP models to perform advanced object detection and visual question answering. This ensemble approach allows users to input images and text prompts, enabling the system to accurately identify objects within the image and answer questions based on the visual content. The tool is designed for tasks requiring detailed visual analysis and contextual understanding, making it suitable for various applications in data labeling and annotation. It is hosted on Hugging Face, providing an accessible platform for users to leverage its capabilities.