georgepearse Goto Github PK

followers: 29.0 following: 76.0 repos: 52.0 gists: 9.0

Name: George Pearse

Type: User

Company: BinIt.ai

Bio: building mini educational ML and analytics apps and writing about them.

Location: London

Blog: https://medium.com/@george.pearse

About Me

I became frustrated enough by the original Detectron2 docs that I decided to start creating my own https://georgepearse.github.io/unofficial-detectron2-docs/.

I write about Machine Learning, Analytics and Start-Ups at https://medium.com/@george.pearse

I studied Physics, then I became a Data Engineering Consultant (Flask + SQL + AWS for orchestration), then I joined a Computer Vision start-up (behold.ai) where I've done a mixed bunch. Set-up the Modern Data Stack but also focused on data-centric-ai via active learning (pytorch-lightning and BAAL) and corrupt sample discovery (cleanlab esk techniques). I've also recently become much more involved in the MLOps community.

Soon to be joining BinIt as Lead ML Engineer. https://binit.ai/

I'm currently working on:

Streamlit app for business intelligence (save SQL for transformations, plotly code for plotting).
Quickdraw app via Streamlit (to try out some computer vision ideas).
Active Learning via bagged ensemble disagreement.
Lightweight UIs for the QDrant Vector Database (and all things QDrant because they build excellent tools) https://github.com/GeorgePearse/QDrant-NLP
Multi-modal ML tools.

Blogs posts

Connect with me:

George Pearse's Projects

activesupervision

Just gonna be a toy / demo built for online active-learning with Supervision.

aim

Aim 💫 — easy-to-use and performant open-source ML experiment tracker.

awesome-open-mlops-data-catalogues

The Fuzzy Labs guide to the universe of open source MLOps

baal-211

BAAL fork to correct for the lack of max defined in np.clip

bagged-disagreement

Active Learning via the disagreement between models trained on overlapping subsets of the data.

bulk-demo

Demo of https://github.com/koaning/bulk on some interesting medical dataset. Ideally in radiology.

cocoframe

Everyone prefers working with DataFrames

colourmodel

Just a tool which uses OpenCV to create bounding boxes around objects of requested colours.

comptoncamera

Geant4 compton camera simulations

data-centric-ai

Resources for Data Centric AI

demo-hnm

diffusers-resize-fix

🤗 Diffusers: State-of-the-art diffusion models for image and audio generation in PyTorch

easy21

Easy21 assignment solution from David Silver's Reinforcement Learning course

fiftyone

The open-source tool for building high-quality datasets and computer vision models

fiftyone-plugins

A curated list of FiftyOne Plugins that you can add to your local FiftyOne install or your FiftyOne Teams deployment

flashcards

gb2fasta

Convert GenBank Files to FASTA Without Leaving the Terminal.

georgepearse

Just for the home page readme

georgepearse.github.io

Place to document independent AI research with a specific focus on Computer Vision and Active Learning

groundingdino

Grounding DINO: Marrying DINO with Grounded Pre-Training for Open-Set Object Detection

groupednms-

Often you want to run NMS over some pairs of classes, but not others which you expect to overlap.

imagecomposer

Package to simplify creating simple synthetic image datasets. You specify the background and foreground objects, the package does the rest.

introspector

Just an idea, hooks implemented for different deep learning frameworks to store the filenames / IDs of the top X losses (or relative losses for that point in training)