Applied AI · Practical guide

How can images become useful research information?

Between an image and a finding are decisions about capture, labels, evaluation and review. Organising them allows models to become part of a research workflow.

Explore the guide
Conceptual schematic. No client data or metrics.

The central idea

The research question defines the visual task.

Computer vision can support image classification, object detection or region segmentation. These are different questions. The appropriate task depends on what researchers want to observe and how results can be checked.

At T2G we start with the task, its data and operating conditions. We agree on a scope that can be checked and explain technical decisions throughout the work.

Step by step

A workflow that preserves traceability.

Four points to organise the work. Each diagram supports the explanation; the process adapts to its context.

  1. 01 / 04

    Define the visual question

    Decide whether the output is a category, object location or a bounded region. Define the unit of analysis: sample, image, field or individual. It shapes capture, annotation and evaluation.

    Category · Location · Region
  2. 02 / 04

    Prepare images with context

    Organise identifiers, capture conditions and labels. Record who annotates and how uncertainty is resolved. Lighting, equipment and preparation can affect models; data should represent the use conditions being evaluated.

    Capture · Context · Labels
  3. 03 / 04

    Evaluate without mixing samples

    Similar images from the same sample should not be split between training and testing without a clear rationale. Split by the unit of analysis and review errors by category and condition. Experts define the reference and its limitations.

    Training · Validation · Test
  4. 04 / 04

    Connect results to their origin

    Save the model version and connect each estimate to its image and sample. Software should support review, correction and export without losing context. Models support analysis; scientific interpretation remains with researchers.

    Estimate · Review · Record

An explained example

Fresh smears: from a visual task to software

01 / Starting point

Recognition and classification require connecting category definitions, models and a way to consult results.

02 / An approach

Within Universidad Santo Tomás’s ANID project, we contributed software engineering, modelling, training and application development. The published case describes that work. These conceptual diagrams show neither samples nor project results.

Explore the published case ↗

For your decision

What needs to be clear.

01

Question

A defined output connected with the research objective.

02

Reference

Expert labelling criteria and recorded disagreements.

03

Traceability

Related images, samples, versions and results.

Common questions

Questions before getting started.

Are classification and detection the same?

No. Classification assigns a category, detection locates objects and segmentation outlines regions. Each needs suitable annotations and evaluation.

Can pretrained models be used?

They may provide a starting point. Their value needs testing on the project’s data and criteria rather than assuming results transfer from another image collection.

Keep exploring

Connect this guide with your challenge.

Let’s discuss your context

Let’s define a useful first step.

We review the problem, available information and constraints with you to assess where it makes sense to start.

Book a diagnostic session directly with the partners