LCIP: Loss-Controlled Inverse Projection of High-Dimensional Image Data

TL;DR

This paper introduces LCIP, a flexible inverse projection method that enables user-controlled exploration of high-dimensional data spaces, improving over fixed-surface limitations for tasks like image style transfer.

Contribution

We propose a novel inverse projection technique that allows sweeping through data space with user control, applicable to any projection method and dataset.

Findings

Supports diverse data augmentation and analysis tasks.

Enables flexible, user-controlled exploration of data space.

Demonstrated effectiveness in image style transfer applications.

Abstract

Projections (or dimensionality reduction) methods $P$ aim to map high-dimensional data to typically 2D scatterplots for visual exploration. Inverse projection methods $P^{-1}$ aim to map this 2D space to the data space to support tasks such as data augmentation, classifier analysis, and data imputation. Current $P^{-1}$ methods suffer from a fundamental limitation -- they can only generate a fixed surface-like structure in data space, which poorly covers the richness of this space. We address this by a new method that can `sweep' the data space under user control. Our method works generically for any $P$ technique and dataset, is controlled by two intuitive user-set parameters, and is simple to implement. We demonstrate it by an extensive application involving image manipulation for style transfer.