Starts: May 23 - 6 pm ET
High resolution drone images provide detailed snapshots of the cities’ landscape. In this working group, we examine the capabilities of multiple ML techniques trained on custom-collected data to develop classification algorithms.
Prior experience using Python is required for this project. We recommend that user’s are also familiar with Google Colab, Tensorflow and Numpy for ML-related modelling tasks. No previous remote sensing knowledge is needed for this project. Please watch the videos in the “Other Useful links + Resources” section to get a clear high level understanding of the whole project: [ https://www.notion.so/aggregate-intellect/7e815b3a5f4e48b3ac6d92fc2c19e7cd?v=aaf9ff169e8e4687916fc560031a8734].
Working Group Leads / Advisors
~ Sophie Nitoslawski | Co-Lead
Drone expert | PhD.c @ University of British Columbia, Canada
~ Ibrahim Elchami | Co-Lead
IoT/AI cloud architect | Postdoc, University of British Columbia, Canada
~ Isabel Todorova | Co-Lead
Drone expert | UBC Graduate @ University of British Columbia, Canada
❗ Meeting Link: Zoom
❗ Meeting Time: Mondays 6-6:30 PM ET on Mondays (starting Monday 23rd of May)
❗ Slack channel [Communication point]: Join by clicking here
❗ Dataset: Drone Images from 3 Parks
❗ Session Slides: Slides [including optional additional information]
❗ Session Videos: Session Videos
- Learn how to access, download and align remote sense imagery from public parks
- Practice organizing and structuring big datasets for streamlined use in custom/computer vision ML projects
- Develop a custom vision ML model using open sources tools, e.g. Yolov4, Yolov5. Model is used to
- [Object detection]: Count trees in a given image
- [Classification]: Classify trees in a given image into groups, such as deciduous or conifer.
- Learn how to quickly demo a ML classifier on images using a webapp (e.g. Flask API).
- Examine model performance and robustness when tested against seen/unseen regions across Canada (and practice implementing hyperparameter optimization)
- Compare model performance against cloud-based tools, e.g. CustomVision.ai from Microsoft.
- Optimize the model (e.g. TFLite, Yolov4-tiny) to run on a smartphone app, simple webapp, and/or run as an edge-AI/embedded-AI component.
- Minimal Project Goal: Develop a ML classifier that can correctly identify and count multiple tree types across different Canadian landscapes
- Stretch Goal:
- Use open-source tools to build a simple, user-friendly webapp makes it user-friendly enough for City employees to process the data in much easier ways.
- Write up a short article that examines this problem and share your findings
Tentative Project Timeline
Expected time to finish
Get familiar with the project domain
Download & process drone data
Develop ML classifiers
Testing / model parameter optimization
Analyze model predictor importances
Build a simple, quick webapp
Our Journey - Drone Analytics-
Aggregate Intellect hosts one of the most diverse ML communities in the world. Over the course of the working group
- Drones are super cool 😎 😎 😎 😎 😎 !!
- You’ll get an immersion into that community & walk out with some cool new friends.
- Advance your ML skills in remote sensing, and how to integrate multi-source data with different uncertainties in Python.
- There are $2500 in gifts budgeted for participants based on contribution. (A nice little push to encourage 😉)
- Contribute to a research study area to better understand the social-ecological interactions in urban environments, which in turn further contributes to building better tools for environmental justice, human health, well-being, and climate change
Looking forward to meeting everyone in our study group! Please feel free to reach out if you have any questions about the planned project.