Azmeraw Bekele Yenew

Azmeraw Bekele Yenew

Auto-generated profile, based on publicly available information

Azmeraw Bekele Yenew

School of Information Technology and Engineering, Addis Ababa Institute of Technology, Addis Ababa University, Addis Ababa, Ethiopia

Research Focus & Metrics

Academic Career Length

3Years (2024-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
29 August 2024 in IEEE Access
Verified
Journal Article
Open access
AYAzmeraw Bekele Yenew
Azmeraw Bekele Yenew
BABeakal Gizachew Assefa
Beakal Gizachew Assefa
EBElefelious Getachew Belay
Elefelious Getachew Belay
Floorplan synthesis is the process of generating new, realistic floor plans for buildings and homes using machine learning and generative models. In recent years, various generative methods, including GANs and diffusion models, have been utilized for the task of floorplan generation, demonstrating promising advancements in architectural design and planning. However, despite their potential, these methods face unique challenges like mode collapse, training instability, and sampling time, which require innovative solutions to overcome for further progress in this field. To address these issues, various techniques such as regualrization techniques, architectural modifications, and optimization algorithms, have been employed. However, existing techniques still struggle to balance both sampling time and diversity simultaneously. In response, HouseGanDi proposes a novel hybrid approach that amalgamates GANs and diffusion models to address the dual challenges of diversity and sampling time in floorplan generation. To the best of our knowledge, this work is the first to introduce a solution that not only balances sampling time and diversity but also enhances the realism of the generated floorplans. HouseGanDi is trained on the RPLAN dataset and combines the advantages of GANs and diffusion models in multimodal fashion while incorporating different techniques such as regularization methods and architectural modifications to optimize our objectives. The multimodality allows our model to jump a number of denoising steps while capturing data distributions. To evaluate the effect of the denoising step, we experimented with different time steps and found better diversity results at T=20. Evaluation of diversity using FID demonstrates an average 15.5% improvement over the state-of-the-art houseDiffusion model, with a 41% reduction in generation time. However, challenges persist in generating non-orthogonal floorplans and accommodating intricate spatial layouts.