Research & Education

Labnote provides a comprehensive research note and data management solution tailored for BT (Bio Tech) and NT (Nano Tech) researchers and organizations. The platform aims to enhance research efficiency by allowing researchers to focus on the core aspects of their work. It features Labnote Scholar, an AI research assistant that helps users maximize their research data, and Labnote Preclindoc, a specialized tool for managing all stages of non-clinical research, including automated report generation. Labnote supports both individual researchers and small to enterprise-level research institutions with tools for data management, research collaboration, and AI-driven insights.

BIOS is an AI Scientist specifically engineered for biological data analysis. This tool has achieved recognition, ranking #1 on BixBench, a benchmark for biological AI. It supports flexible workflows, allowing for human-in-the-loop checkpoints to ensure oversight and control, alongside a fully autonomous operational mode. The system leverages specialized AI agents, each designed for distinct tasks such as orchestration, comprehensive literature review, in-depth data analysis, and advanced novelty detection within biological datasets. A free tier is available, with academic users benefiting from full, complimentary access using their .edu email addresses.

ChemBench Leaderboard is an AI tool designed to benchmark and compare the performance of various AI models in chemistry-related tasks. Hosted on Hugging Face Spaces, it offers a user-friendly interface to browse a searchable and filterable leaderboard of models, displaying their performance scores across different metrics. Users can customize which columns to display, making it easy to focus on relevant data. The platform also provides functionality for users to upload their own model's evaluation results, contributing to the community and expanding the dataset for comparison. Built with Gradio, this open-source tool is available for free under the MIT license, promoting transparency and collaboration in scientific AI research.

Cellpose is a generalist AI algorithm designed for cellular segmentation, applicable across various cell types and imaging modalities. Users can upload image files such as PNG, JPG, or TIF, and the application will process them to generate precise outlines of cells. Beyond static segmentation, Cellpose also provides flow images, which are useful for visualizing and analyzing cell movement. This tool is built using Gradio and is available under the BSD-3-Clause-Clear license, making it accessible for a wide range of research and analytical purposes in biology and related fields.

Efficient Teacher, developed by Alibaba, is a comprehensive open-source library designed for both supervised and semi-supervised object detection (SSOD) using the YOLO series. Built upon the YOLOv5 framework, it leverages YACS and advanced network designs to restructure key modules, enabling a single algorithm library to support training for YOLOv5, YOLOX, YOLOv6, YOLOv7, and YOLOv8. This tool is particularly beneficial for scenarios with domain differences between training and deployment, high data labeling costs, or limited labeled data. It introduces semi-supervised object detection into practical applications, allowing users to achieve strong generalization capabilities with a small amount of labeled data and a large amount of unlabeled data. Efficient Teacher also provides features like category and custom uniform sampling to quickly improve network performance in business scenarios. It offers scripts to convert YOLOv5 weights, use existing YOLOv5 datasets without format adjustments, and easily switch between different YOLO network structures via YAML configuration.

EpipolarPose is a PyTorch implementation for self-supervised learning of 3D human pose using multi-view geometry, as presented in the CVPR 2019 paper. This tool is designed for computer vision researchers to estimate 3D human poses without the need for extensive 3D ground-truth data or camera extrinsics during training. It works by estimating 2D poses from multi-view images and then leveraging epipolar geometry to derive 3D poses and camera geometry, which are subsequently used to train a 3D pose estimator. In the testing phase, it can produce a 3D pose result from a single RGB image. The project includes scripts for training and validation, data preparation utilities, and pre-trained models on datasets like Human3.6M and MPII.

FacePose_pytorch provides a PyTorch implementation for real-time head pose estimation (yaw, roll, pitch) and emotion detection, boasting state-of-the-art performance. The tool is designed for easy deployment and use, offering high accuracy in solving various face detection problems. It utilizes Retinaface for face frame extraction, PFLD for key point identification, and a simple linear model for pose estimation. Additionally, it incorporates a highly accurate emotion recognition model, achieving impressive results on datasets like raf-db, affectnet, and ferplus, predicting seven types of expressions. The project emphasizes its efficiency and accuracy compared to existing open-source solutions.

DeepLabCut Model Zoo is a specialized tool designed for animal pose estimation, hosted on Hugging Face. It enables users to upload images and apply pre-trained models to detect animals and estimate their poses. The application offers a selection of animal detectors and pose-estimation models, drawing bounding boxes and keypoint markers on identified animals. Users can also adjust confidence thresholds for more precise results. This tool is particularly useful for researchers and scientists in fields requiring detailed analysis of animal behavior and movement tracking.

DINOv3 is an AI tool designed for advanced image analysis, specifically focusing on similarity and classification tasks. Users can upload multiple images to the platform to compute their cosine similarity, which helps in identifying visually similar content. Beyond similarity analysis, DINOv3 enables users to build custom classifiers by adding images to different categories. This functionality allows for the prediction of classes for new, unseen images, making it a versatile tool for various computer vision applications. It is particularly useful for researchers and developers who need to analyze and categorize large datasets of images efficiently.

DINOv3 Keypoint Matching is an AI tool hosted on Hugging Face Spaces, designed to identify and highlight corresponding keypoints across two uploaded images. Users can leverage various DINOv3 models to optimize the accuracy of keypoint detection and matching. This tool is particularly useful for tasks requiring precise visual correspondence, such as object recognition, image analysis, and computer vision research. Its web-based interface makes it accessible for quick experimentation and demonstration of DINOv3's capabilities in visual feature extraction and matching.

Depth Compare is an AI tool designed for comparing various depth estimation models. Built with Gradio, it provides a platform for users to evaluate the accuracy and performance of different depth maps. The application checks for and installs necessary dependencies like Pixi and Homebrew, manages processes on port 7860, and runs within a Pixi application environment. While the current live website indicates a runtime error, the tool's intent is to facilitate research and educational purposes by offering a comparative analysis of depth estimation techniques.

Depth Estimation is an AI tool designed to estimate depth from images, providing a visual representation of depth information. Built with Gradio, it offers a user-friendly interface for generating depth maps from various visual inputs. This tool is particularly useful for researchers, developers, and students in the fields of AI and computer vision, enabling them to explore and apply depth estimation techniques. While the current live website indicates a runtime error, the underlying functionality aims to provide a practical application for understanding spatial relationships within images.

Depth Anything is an AI tool available on Hugging Face that specializes in depth estimation from single images. Users can upload an image, and the application processes it to estimate the distance of each element within the scene. The output is a colored depth map, which provides a visual representation of the inferred depth information. An interactive slider allows for easy comparison between the original image and the generated depth map. Additionally, the tool provides a 16-bit raw depth output, catering to more advanced applications. This capability is valuable for various fields, including 3D scene understanding, robotics, and computer vision research.

Depth Anything V1 vs V2 is a specialized tool designed for researchers and developers in the field of computer vision and depth estimation. It provides a direct comparison between two versions of the Depth Anything model, allowing users to upload an image and visualize the generated depth maps from both V1 and V2 simultaneously. This side-by-side comparison is invaluable for understanding the improvements, differences, and performance characteristics of each model. Users can also select different model sizes for each version, offering flexibility in evaluating the trade-offs between accuracy and computational cost. The tool serves as an excellent resource for analyzing and improving depth estimation algorithms.

Depth Anything V2 is an advanced AI tool designed for estimating depth from single images, offering improved performance over its predecessor. Hosted on Hugging Face Spaces, this application allows users to upload an image and receive a comprehensive depth map. The output includes a vibrant, colorful depth visualization, a clear grayscale depth image, and a 16-bit depth map, providing versatile data for various applications. Built with Gradio for an intuitive user interface, it simplifies the process of obtaining detailed depth information, making it accessible for researchers, developers, and anyone interested in computer vision tasks.

Distill Any Depth is an AI tool designed for monocular depth estimation, allowing users to upload any picture and receive an estimate of how far each part of the scene is. The application utilizes knowledge distillation algorithms to create detailed depth maps from single images. It provides a colorful depth image that can be explored with a slider, a plain grayscale depth view for a more traditional representation, and a downloadable raw depth map for further analysis. This tool is particularly useful for computer vision research and applications requiring precise depth information from 2D images. It is available under the Apache 2.0 license.

Dpt Depth Estimation + 3D Voxels is an AI tool available as a Hugging Face Space that allows users to upload an image and generate a corresponding depth map. From this depth map, the tool reconstructs a 3D voxel model, providing a three-dimensional representation of the input image. A key feature is the ability to adjust the voxel size, which directly influences the level of detail in the resulting 3D model. This functionality makes it suitable for exploring 3D reconstruction from 2D images, catering to individuals interested in computer vision, 3D modeling, or experimental AI applications.

E2E FT GeoWizard is a Hugging Face Space that provides end-to-end fine-tuned monocular depth and normal estimation from images. Users can easily upload an image to the platform, select their desired processing resolution, and then generate detailed depth and normal maps. The tool supports downloading the generated maps in various formats, making it versatile for different applications. It is designed for in-the-wild, zero-shot, single-step depth analysis, offering a straightforward solution for visual data processing. The tool is licensed under Apache-2.0, indicating its open-source nature and potential for community contributions.

Dpt Depth Estimation is an AI tool hosted on Hugging Face Spaces, designed to generate depth maps from uploaded images. This application processes an input image and outputs a visual representation of depth, where the brightness of objects indicates their distance from the viewer—brighter objects are closer. It leverages the Dpt model for accurate depth estimation, making it a valuable resource for various computer vision tasks. The tool is straightforward to use, requiring only an image upload to produce the depth map, making it accessible for quick analysis and visualization.

The FAIR Chemistry Leaderboard is a Hugging Face Space developed by Facebook for chemistry research. It enables researchers to upload their model prediction files, such as NPZ or JSON, along with essential information about their model and training dataset. The platform then evaluates these predictions against established reference data, providing a standardized way to track and compare model performance. This tool is designed to foster progress in chemistry-related tasks by offering a transparent and collaborative environment for benchmarking AI models in the field. It is built with Gradio and requires Hugging Face authentication for access.

Feat2GS is an AI tool hosted on Hugging Face Spaces, designed for generating 3D models from a series of input images. Users can upload multiple images of a scene, and the application will process them to extract relevant features. Following feature extraction, Feat2GS optimizes the 3D model, ensuring a high-quality representation of the scene. Finally, it renders the generated 3D model into a video, allowing users to select a specific camera trajectory for the output. This tool is built using Gradio and Python, and it operates as a web application, making it accessible for various users. It is licensed under Apache-2.0, indicating its open-source nature.

Generating molecular graphs by WGAN-GP is an AI tool hosted on Hugging Face Spaces, designed to create molecular graphs. The tool leverages a WGAN-GP (Wasserstein Generative Adversarial Network with Gradient Penalty) model for its generative capabilities. While the concept aims to assist in the design of new molecules, potentially benefiting chemists and materials scientists, the current live website indicates a runtime error, preventing its functionality. The error message suggests an issue with loading the Keras model, specifically regarding unsupported file formats in Keras 3. This indicates a technical challenge that needs resolution for the tool to become operational.

MARS (Modular and Realistic Simulator for Autonomous Driving) is an open-source project designed to provide an instance-aware, modular, and realistic simulation environment for autonomous driving research. It allows users to train and test autonomous vehicle algorithms using various datasets like KITTI and vKITTI2. The simulator supports reconstruction and novel view synthesis tasks, offering pre-trained models and the flexibility to train from scratch with custom data. Its modular framework enables combining different architectures for various nodes, such as using Nerfacto for background models. MARS is built upon Nerfstudio and requires an NVIDIA GPU with CUDA for installation and operation.

[navhard] NAVSIM v2 End-to-End Driving offers an AI simulation environment specifically designed for autonomous vehicle research. This platform enables users to view competition details, access relevant datasets, and check leaderboards to benchmark their end-to-end driving models. Researchers and developers can manage their submissions and review submission information, fostering a competitive and collaborative environment for advancing autonomous driving technology. The tool is hosted as a Hugging Face Space, indicating its accessibility and potential for community engagement in the field of AI-driven vehicle simulation.