All-sky Image Segmentation on Edge Compute for Wx Labs

by Wx Labs Ltd in
17 (views)

Project insights

  • Location

    United Kingdom

  • Open to

    2 Student

Description

1. Context

WxStation captures all-sky images every 5 minutes. Wx Labs would like to explore running a segmentation model locally (on the device) to classify pixels into cloud types, contrails and ground. This project is to build a first working version of that pipeline, and then optimise it to run on the target hardware.

2. Aim

To explore and evaluate approaches to image segmentation under constrained hardware conditions, with a focus on understanding methodological trade-offs, design considerations, and feasibility in low-resource environments.
This project is exploratory and educational in nature. It is not intended to produce production-ready systems, commercially deployable outputs, or components for integration into any live or planned products.

3. What the Student Practice

The student will independently design and conduct a structured investigation into segmentation approaches. Activities are framed as experiments and analysis, not the delivery of business solutions.
The student may:
  • Define and justify a dataset structuring and labelling approach (e.g. pixels vs polygons), including documenting assumptions and limitations
  • Explore alternative labelling schemas, including:
    • a fine-grained experimental schema
    • a simplified schema for comparative analysis
  • Design an experimental configuration to test how variations in:
    • class structure
    • input resolution
    • model architecture affect observed outcomes
  • Conduct controlled experiments using a baseline segmentation approach to:
    • observe performance characteristics
    • analyse trade-offs between different configurations
    • identify limitations and uncertainties
  • Develop non-production, illustrative prototypes to demonstrate conceptual inference workflows, including:
    • simulated handling of inputs
    • indicative outputs (e.g. segmentation masks and summary data)
    • high-level observations on expected runtime and memory behaviour
  • Investigate optimisation strategies (e.g. compression, quantisation, class simplification) at a conceptual and experimental level only, without expectation of deployable optimisation
  • Produce a constrained environment analysis that discusses:
    • estimated computational requirements
    • trade-offs between efficiency and accuracy
    • theoretical feasibility of deployment on devices such as Raspberry Pi
All activities are undertaken as part of a learning exercise. The student is not responsible for delivering outputs that meet business, commercial, or operational requirements.

4. What Wx Labs will Support

Wx Labs will act in a supportive, non-directive capacity and will not supervise the student as an employee or contractor.
Wx Labs may:
  • Provide example or synthetic datasets for experimentation (where possible, non-sensitive and non-production data)
  • Offer high-level contextual information about general constraints relevant to edge environments
  • Provide periodic, non-binding feedback on the student’s ideas and interpretations
  • Make hardware available optionally, strictly for exploratory testing, with no requirement to use it
Wx Labs will not:
    • Require specific deliverables for business use
    • assign tasks, deadlines, or performance expectations
    • integrate the student into internal workflows, teams, or product development processes
    • relies on the student’s work for commercial or operational purposes

5. Deliverables

The outputs are academic-style and demonstrative, intended to evidence learning and exploration rather than practical deployment. The student will produce:
A standalone repository containing:
  • Experimental code used to explore different approaches
  • A configurable framework for testing hypotheses (research-oriented, not production-ready)
  • Demonstration scripts illustrating concepts (non-production, non-integrated)
A written report covering:
  • Dataset design decisions and labelling rationale
  • Comparative analysis of segmentation approaches
  • Summary of experimental findings and observed trade-offs
  • Conceptual discussion of optimisation strategies
  • Feasibility analysis for constrained environments
  • Limitations of the investigation and areas for further exploration
6. Boundary Statement
This project is:
  • Not employment, work, or provision of services to Wx Labs
  • Not intended to generate commercially usable outputs
  • Not part of Wx Labs’ product development or delivery pipeline

Interested in this project?

39 days left to apply

Apply for this project

Cancel
Send message
Cancel