Find open source projects actively accepting contributors. Search repositories, filter by program milestones, difficulty tags, or tech stack.
Use our Orbit AI Matcher to find out! Get instant matching scores based on your developer skills, preferred frameworks, and contribution experience.
Convert your selected open-source project into a winning GSoC, LFX, or Outreachy application using Proposal Studio.
<p>TensorFlow (TF) and PyTorch are some of the most popular deep learning frameworks. TF exports networks in its own format; PyTorch supports export of the trained models into ONNX format, which is also very popular. OpenCV is becoming quite popular tool for Deep Learning Inference. It would be very useful to describe clearly how to train a model in TF or PyTorch so that it can be then run in OpenCV.</p>
This project enables global forecasting in sktime. 1. design and implement global forecasting API in sktime. 2. add pytorch-forecasting models with global forecasting API. 3. extend HFTransformersForecaster, NeuralForecast models to global mode. 4. extend forecasting benchmark and NaiveForecaster to global mode. 5. documents and notebooks for global forecasting in sktime.
Current Mariadb ColumnStore engine doesn’t support vectorized evaluation of SQL functions and expressions. This project want to cover this by adding row/col transformation before the evaluation and evaluate the performance improvement. Furthermore, this project will also try to accomplish a delayed projection into RowGroup, which store data in vertical stored buffers and optimize the transformation part.
<p>The objective of this project is to create an open-source plugin for SOFA building an interface with the open-source library Pulse. This engine provides comprehensive models for the human physiology.</p> <p>The coupling will lead to an interface which will model parts of the human physiology in addition to providing simulation data of various components, varying with time.</p>
<p>The current parallelism model of GeantV is data-oriented where a static set of threads is predefined to perform the work, the set of transport threads and distributor thread. In this model a concurrent queue data structure is used as a common source for threads. The main objective of this project is to adapt the steering of this work to a task-based approach using Thread Building Blocks (TBB).</p>
The aim of this project is to create healthcare examples using Keras. I plan to implement new tutorials for: 1. Diabetic Retinopathy detection using SOTA EfficientNet model. 2. Pulmonary disease diagnosis using cough sounds by creating a feature extraction pipeline and a Deep CNN model. 3. Skin lesion segmentation using Semi-supervised contrastive learning.
This project uses OpenCV to identify a track for an autonomous vehicle to follow. A new model using EfficientDet along with an existing training set of images building upon previous work is proposed. The model will be ported to TFlite and implemented on the Coral USB Accelerator. An evaluation of its performance against previous efforts will be performed. Additionally, mapping landmarks exploiting known geometry is performed.
This project will focus on applying hybrid end-to-end models that combine weighted convolutional-recurrent architectures (CNN-RNN) with large language models (LLMs) to recognize text from Spanish printed sources dating back to the seventeenth century. The ultimate goal is to create a user-friendly application that allows researchers to generate transcripts simply by uploading images in PDF or PNG format.
<p>A web app that does automatic labeling of radiology images. The backend is in Django and tensorflow-serving whereas the frontend is in Reactjs. The web app will help make dataset bigger and better in radiology domain. The user will be able to visualise deep learning model output such as bounding box and segmentation and modify them in real time to enlarge the dataset and intern further improve the DL models.</p>
<p>TensorFlow Hub and Model Garden are the go-to resources for ML practitioners for quick and easy experimentation. This project aims to implement <strong>RegNet-Y</strong> from the paper "Designing Network Design Spaces" by Radosavovic et. al and publish the pre-trained models on TF-Hub. RegNet-Y provides improved inference latency and it can be scaled according to the requirements of the user.</p>
The following points will be completed by the end of the projects: ● Data collection. ● Data preprocessing ● Choosing and implementing an AI solution and training the AI model on the preprocessed data ● Testing and iterative refinement of the model ● Integration and deployment of the AI system (e.g., Slack channel, website chatbot) ● Documentation and user guidance The final product will be an AI-powered support system for the Mifos/Fineract community
The main target of this project would be to work on migration to user groups for permissions which aims to replace every setting in Zulip that currently allows organizations to assign permissions based on role (admin, moderator, etc.) with a setting based on arbitrary “user groups”. I would majorly work on areas of settings and stream settings, migrating the existing settings to the new model and adding new settings based on the new permission model.
<p>The idea of this project is to extend the existing JOSM Plugin "indoorHelper" by an import feature for Building Information Modelling (BIM) files. With this import feature it will be possible to convert BIM modelled indoor data to OSM. See also <a href="https://wiki.openstreetmap.org/wiki/Google_Summer_of_Code/2020/BIM_Import" target="_blank">https://wiki.openstreetmap.org/wiki/Google_Summer_of_Code/2020/BIM_Import</a>.</p>
<p>The project aims to provide a compiler extension and runtime for a new "paramserv" built-in function. In the context of large-scale machine learning, it will bring more performance when training a model with a data-parallel and model-parallel parameter server. SystemML already supports the data-parallel and task-parallel operator. Hence, we can focus on designing the parameter server primitive and implementing the parameter update strategies.</p>
<p>Developing a tool for modeling Newtonian physics processes based on the LibGeoDecomp library and using HPX for parallel execution. Tool should be easy for use, it should enable end users the creation of simulations with a minimal amount of code. (user can model scene in Blender, export it in text format and pass as argument to the command). The user uses domain terms, rather than writing a large number of irrelevant code.</p>
<p>FluxJS is a tool to export Flux models to javascript using deeplearn.js. Since Flux is a new library, it is essential to have lots of examples demonstrating its use. This would make it easy for new users to pick it up. Along with the great features of flux, its website should showcase good demos. It is important that the demos be appealing to clients. Hence, the projects listed include applying AI on visual art and games through Flux models.</p>
The major objectives of the project are: 1. Refactoring the Python code to increase efficiency and readability and increase the overall code quality of the entire codebase of the software. 2. To enhance user engagement and support by creating a comprehensive FAQ dataset which is used to train a custom Chatbot. 3. Investigate and discuss different possibilities for multi-GPU model execution, i.e. allow a model to execute (and share memory) across multiple GPUs.
<p>DeepFalcon is an ultra-fast non-parametric detector simulation package. This project aims to extend DeepFalcon by adding functionality for Graph Normalizing Flows (GNFs) to it.</p> <p>While previous work on DeepFalcon has focused on other Deep Generative Models and Graph Neural Networks to simulate detector event reconstructions, we aim to use GNFs to create generative models for graph structures that can better approximate detector reconstructions.</p>
PyMC uses PyTensor as its computational backend, which allows efficient optimisation, rewrites and evaluation of mathematical operations. Using a recently introduced framework (CoLa- Compositional Linear Algebra), we identify a lot of algebraic simplifications and implement them as graph rewrites. As a result, we would be speeding up multiple model families in PyMC, such as Time Series, State-Space models and Gaussian Processes. The goal is to simplify efforts in implementations while achieving a high degree of computational efficiency
<p>At present caMicroscope has many features like marking up images, running machine learning models on individual images, getting the predictions and downloading the image parts . The aim of the project is to extend these features and add new ones such as letting the users fetch regions of interest automatically and store them along with their labels , supporting a wider range of models and having a feature for normalization of histological stains / synthetic data generation of cancer tissue images .</p>
MLModelScope (https://mlmodelscope.org/) is a playground that allows people to try ML tasks with various model types, frameworks and hardware backends. It is jointly developed by IBM and University of Illinois. It is currently one of the platforms that collaborate with MLPerf for ML benchmark workload validation. (IBM participates in MLPerf as a founding members, including contribution from z). MLModelScope enables multiple backends for model training, execution and performance study. Currently IBM Z is enabled.
This project intends to develop cutting-edge machine-learning tools for spectral analysis to characterize the atmospheres of exoplanets. The project will explore various state-of-the-art and recent techniques to build a robust and general model with strong regressional properties. The resultant models from our study will be benchmarked against pre-existing baselines based on the principles of statistics and atmospheric sciences. Additionally, relevant tutorials and documentation will be provided, keeping future developments in mind.
<p>A tool for producing interactive visualizations from data generated by the <a href="https://github.com/moja-global/About_moja_global/wiki/GCBM-is-runs-CBM-science-models-on-the-FLINT-platform" target="_blank">Generic Carbon Budget Model</a> (GCBM).</p> <p>For the <a href="https://moja.global" target="_blank">moja global</a> sub-org, this project is an experiment that aims to explore the dynamic ways in which GCBM output can be viewed.</p>
<p>PODIO is a C++ library that allows the creation of event data models and efficient I/O code for HEP experiments. It does so by avoiding deep-object hierarchies and virtual inheritance. On the other hand, HDF5 is a data format that allows one to manage extremely large and complex data collections. Due to its versatility as a data model and rich set of performance features, it is an ideal format to store PODIO data in. The aim of this project is to implement an HDF5 backend for PODIO.</p>