Related Experiment Video
Updated: Aug 22, 2025
08:25
Combining Eye-tracking Data with an Analysis of Video Content from Free-viewing a Video of a Walk in an Urban Park Environment
Published on: May 7, 2019
9.1K
Exploration of Semantic Label Decomposition and Dataset Size in Semantic Indoor Scenes Synthesis via Optimized
Hatem Ibrahem1, Ahmed Salem1,2, Hyun-Soo Kang1
1Department of Information and Communication Engineering, School of Electrical and Computer Engineering, Chungbuk National University, Cheongju-si 28644, Korea.
Sensors (Basel, Switzerland)
|November 11, 2022
Summary
This study enhances Pix2Pix for realistic indoor scene generation using optimized architectures. The proposed residual connections-based models offer improved performance with fewer parameters and better image quality compared to existing methods.
Area of Science:
- Computer Vision
- Artificial Intelligence
- Machine Learning
Background:
- Paired image-to-image translation is crucial for generating realistic visual content.
- Conditional Generative Adversarial Networks (Pix2Pix) are a key architecture in this domain.
- Optimizing Pix2Pix for indoor scene generation requires architectural and training enhancements.
Purpose of the Study:
- To propose efficient optimization techniques for Pix2Pix architecture and training.
- To enhance the realism of generated indoor scenes using semantic segmentation maps.
- To develop a generative adversarial network-based technique for artificial indoor scene creation.
Main Methods:
- Implementing residual connections in generator and discriminator architectures.
- Training models on the NYU depth-v2 and ADE20K indoor datasets.
- Comparing proposed models against state-of-the-art methods using LPIPS and FID metrics.
Main Results:
- Proposed models demonstrate fewer parameters and less computational complexity.
- Generated images exhibit superior quality compared to Pix2Pix and other recent methods.
- Residual connections-based models show better learning from small datasets and improved realism on larger ones.
Conclusions:
- The proposed optimization techniques significantly boost the performance of Pix2Pix for indoor scene generation.
- Residual connections-based models offer a more efficient and effective approach to realistic image synthesis.
- The method achieves state-of-the-art results in terms of image realism and translation quality.
