I collected virtual food models from 3dsky. Although these models resemble real food, they exist as digital assets. I wanted to explore how they could be catalogued using a different system.

I selected all the food models on the last page of the Food & Drinks section on 3dsky, collecting 28 models in total.

Method 1 — Reclassification
Grouping 28 food models by sugar and fat content.

Method 2 — (Re)presenting
Reorganising models by colour labels.

Method 3 — Taxonomising
Exploring overlapping categories among 28 models.

Further Development
I chose to develop Method 1 further, exploring how the new classification could happen.

Why Reclassify?

The original website mainly organises food models by their names and types.

I would like to try a different approach: What would happen if we classify foods based on their compositional data, rather than their names?

Different classification criteria make certain relationships more apparent, while other relationships remain hidden.

For example, Banana and Sushi are originally different types of food, but in my system, they are grouped together into the same category.

This shows that classification isn’t just about organizing items; it also changes the relationships between those items that are visible.

How？

Since 3D models themselves do not contain any nutritional values, I found reference data for corresponding real foods for each model. Using the sugar and fat content per 100 grams, new classification rules were established.

I assigned reference values for sugar and fat content to each of the 28 food items, based on the actual nutritional data of those foods. Then, I divided these 28 models into four categories according to two thresholds.

I used 20g of sugar and 8.5g of fat per 100g as the dividing lines, creating four categories.

I turned each model into an individual archive record. Each card contains its image, identification number, reference values and new classification.

Why are cards necessary?

Charts primarily illustrate the relationships between data points, while archive cards enable each model to serve as a record that can be searched for, accessed, and rearranged independently.

What matters here is the transition from data visualization to actual cataloging operations – not how beautiful the cards are.

A New Retrieval System

I created a master index listing all 28 models by category, with their IDs and reference values. The index connects the classification system to the individual archive cards.

The master index allows users to locate a model by category and ID, then find its corresponding archive card.

Reflection

While organising the 28 models, I noticed that some classification results depended heavily on the chosen thresholds.

For example, Cookies contains 19.90g of reference sugar per 100g, placing it in the low-sugar category. An increase of just 0.10g would move it into a different category.

This made me question how much the classification depends on the rules I set, rather than the models themselves.

Even without changing the data, moving the dividing line could change which category a model belongs to.

How can small changes in data affect the way we classify things?