Project Page -
Harvesting natural points-of-view preferences for arbitrary objects

OpenSource framework for harvesting point-of-view preferences silently captured while users perform unrelated visual tasks.

Gianluca Roveda1, Georgios Papaioannou2, Andreas A. Vasilakis2, Marco Tarini1

1 University of Milano, Italy  ·  2 Athens University of Economics and Business, Greece

An earlier version of this project won the STAG 2025 Thesis Award.

Download the Dataset Harvesting Tool

Harvesting Natural Points-of-View Preferences for Arbitrary Objects

In press · 3DOR 2026 — Eurographics Symposium on 3D Object Retrieval

Paper preprint

Download
A peek of our dataset: left, a cloud of viewpoints around a chair model; right, their clustrization
A peek of our dataset: left, a cloud of viewpoints around a chair model; right, their clustrization

ABOUT

Finding the optimal viewpoint for a 3D object is crucial for visualization, digital marketing, and shape retrieval. Because viewpoint preference is driven by subjective semantic and aesthetic biases, data-driven approaches require large datasets of user preferences. This project presents a framework to collect viewpoint data by implicitly tracking users during casual interactions with 3D objects. We publicly release both the dataset and the configurable tool used to gather it.

DATASET & TOOL

Code & Data