Add a minimal non-frontend test workflow for the Interactive Classifier

[kunfang98927]

IC end-to-end workflow validation without a frontend

Author: Kun
Reviewer: Kyrie
Approver: The Team
Status: Draft
Date: 2026-05-20
Related: DDMAL/Standalone-Interactive-Classifier

Context

We currently cannot run or test the full IC workflow end-to-end because there is no frontend. Before building a UI, we need a lightweight way to validate the workflow using the known three inputs and see whether the workflow can generate the expected two outputs.

Problem statement

Right now, there is no simple, repeatable way to run the IC workflow from input to output generation. This makes it hard to confirm that the workflow is runnable and bug-free. More importantly, it's hard to compare different feature extraction methods without the whole workflow running.

Goals

• enable end-to-end IC workflow without a frontend
• validate the three inputs and two outputs
• make the workflow easy to rerun for debugging
• support simple interactive process

Non-goals

• building the frontend
• finalizing the UI or API design

Proposed approach

Build a small script-based test harness for IC.

It should take:

1. a full image
2. a bounding-box file
3. a class-label file

and produce:

1. predicted labels for each bounding box
2. a structured export xml file

Example:

python run_ic_process.py --image sample.png --boxes boxes.json --labels labels.json --output out.xml

Optionally, it can also generate an annotated image for visualization.

Alternatives considered

• Build a frontend: more intuitive, but slower and harder to debug at this stage.

Risks and open questions

• Risk: a structured output file alone may not be easy to validate by eye.
• Question: we should decide whether visualization is required for the first version or can be added later.

Success criteria

• the IC workflow can be run locally without a frontend
• the script accepts the three required inputs and produces the two expected outputs
• at least one sample case can be rerun reliably

Rough plan

• Define input and output formats
• Implement a script to do single IC workflow (input->output)
• Decide whether to include visualization (e.g., mark the model outputs on the original image?)
• Add a minimal multi-round interactive workflow (e.g., input->output->manual modification->output)
• Document setup and usage for this whole workflow

[kunfang98927]

I'll upload the above things as a design doc and place it into agile repo.

[kunfang98927]

I'll upload the above things as a design doc and place it into agile repo.

Already uploaded and got approved here: https://github.com/DDMAL/agile/blob/main/design-docs/mothra/Mothra_IC-workflow-validation_designdoc_20may2026.md