All case studies

Computer visionResearch portfolio project

Explore mammal identification with computer vision

Mammal World brings species and behaviour labels into a computer vision experiment, with dataset preparation, model training and an inference demonstration.

Contribution
Mehrab Mashrafi
Scope
Dataset curation, transfer learning and model demonstration
Project context
Original portfolio experiment

Documented dataset coverage

103 species

Scope of collected imagery, not complete regional coverage or field accuracy.

Build the label set before training

Mammal records from the India Biodiversity Portal were checked against other reference sources. Images were collected and cleaned for the selected species.

Mehrab explored whether species and behaviour labels could be handled together while keeping the preparation steps inspectable.

Distinguish a prediction from an observation

A transfer learning experiment produced behaviour predictions that were incorporated into the later dataset and model workflow.

Those predictions are not expert confirmed observations. A user needs to distinguish a reference label from one supplied by another model before relying on it.

How the work connects

  1. 01

    Species references

  2. 02

    Curated images

  3. 03

    Behaviour labels

  4. 04

    Model training

  5. 05

    Predictions

What this work establishes

The project documents imagery for 103 mammal species and explores species and behaviour predictions in a shared workflow.

This documents a model experiment, not a wildlife monitoring deployment. Conservation outcomes and performance on independent field imagery are not established.

Inspect the work or discuss the approach

Have a similar workflow to improve?

Tell us who uses it, which records it depends on and where the handoff slows down. We can discuss a practical implementation scope.

Discuss your use case

Project notes from Mehrab Mashrafi and AI Byters. Updated .