Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Adversarial Latent Autoencoder with Self-Attention for Structural Image Synthesis

Jiajie Fan, Laure Vuaille, Hao Wang, Thomas Bäck

TL;DR

This work proposes the Self-Attention Adversarial Latent Autoencoder (SA-ALAE), which allows for generating realistic structure designs of complex engineering parts and showcases the potential of SA-ALAE by generating engineering blueprints in a real automotive design task.

Abstract

Generative Engineering Design approaches driven by Deep Generative Models (DGM) have been proposed to facilitate industrial engineering processes. In such processes, designs often come in the form of images, such as blueprints, engineering drawings, and CAD models depending on the level of detail. DGMs have been successfully employed for synthesis of natural images, e.g., displaying animals, human faces and landscapes. However, industrial design images are fundamentally different from natural scenes in that they contain rich structural patterns and long-range dependencies, which are challenging for convolution-based DGMs to generate. Moreover, DGM-driven generation process is typically triggered based on random noisy inputs, which outputs unpredictable samples and thus cannot perform an efficient industrial design exploration. We tackle these challenges by proposing a novel model Self-Attention Adversarial Latent Autoencoder (SA-ALAE), which allows generating feasible design images of complex engineering parts. With SA-ALAE, users can not only explore novel variants of an existing design, but also control the generation process by operating in latent space. The potential of SA-ALAE is shown by generating engineering blueprints in a real automotive design task.

Adversarial Latent Autoencoder with Self-Attention for Structural Image Synthesis

TL;DR

This work proposes the Self-Attention Adversarial Latent Autoencoder (SA-ALAE), which allows for generating realistic structure designs of complex engineering parts and showcases the potential of SA-ALAE by generating engineering blueprints in a real automotive design task.

Abstract

Generative Engineering Design approaches driven by Deep Generative Models (DGM) have been proposed to facilitate industrial engineering processes. In such processes, designs often come in the form of images, such as blueprints, engineering drawings, and CAD models depending on the level of detail. DGMs have been successfully employed for synthesis of natural images, e.g., displaying animals, human faces and landscapes. However, industrial design images are fundamentally different from natural scenes in that they contain rich structural patterns and long-range dependencies, which are challenging for convolution-based DGMs to generate. Moreover, DGM-driven generation process is typically triggered based on random noisy inputs, which outputs unpredictable samples and thus cannot perform an efficient industrial design exploration. We tackle these challenges by proposing a novel model Self-Attention Adversarial Latent Autoencoder (SA-ALAE), which allows generating feasible design images of complex engineering parts. With SA-ALAE, users can not only explore novel variants of an existing design, but also control the generation process by operating in latent space. The potential of SA-ALAE is shown by generating engineering blueprints in a real automotive design task.
Paper Structure (13 sections, 5 equations, 7 figures, 1 table)

This paper contains 13 sections, 5 equations, 7 figures, 1 table.

Table of Contents

  1. Introduction
  2. Related work
  3. Methods
  4. Background
  5. Improved ALAE
  6. Self-Attention Adversarial Latent Autoencoder (SA-ALAE)
  7. Experiments
  8. Datasets
  9. Evaluation methods
  10. Training configurations
  11. Results
  12. Interpolation in Latent Space
  13. Conclusion

Figures (7)

  • Figure 1: Vehicle A-Pillar 3D model (left) and various versions of A-Pillar blueprints obtained by extracting cross-sections from the 3D object.
  • Figure 2: SA-ALAE in directed graphs. ALAE and SA-ALAE share an identical training strategy. The sampling procedure of SA-ALAE utilizes the off-the-shelf mapper $M$ to sample additional noisy latent variables $\omega_{noise}$, hereby introducing randomness in the sampling.
  • Figure 3: FID measured in terms of sample size. TL-0,1,2, and 3 stand for various target locations on vehicle A-Pillars where the cross-sections are taken. For this experiment, all blueprints are from Dataset$_1$.
  • Figure 4: Examples of source blueprints from Dataset$_1$ and generated blueprints.
  • Figure 5: Examples of source blueprints of A-Pillar from Dataset$_1$ and blueprints generated by SA-ALAE$_{sub}$.
  • ...and 2 more figures