Disaster Simulacra

Disaster Simulacra

Introduction

In the face of our rapidly changing climate, the need for heightened awareness and understanding has never been more crucial. Disaster Simulacra is not merely a  disaster & extreme environment visualizer but a call to action—a resonant echo of our commitment to visualizing the impact of disasters and extreme environments on familiar surroundings. It encapsulates the essence of our generative AI initiative, which seeks to bridge the gap between the abstract notion of climate change and its tangible consequences. This is project is collaborated with Angelina Zhang and I was mainly responsible for the technology implementation.

Screenshot 2024-01-10 at 7.25.14 PMScreenshot 2024-01-10 at 7.25.20 PM
Screenshot 2024-01-10 at 7.25.14 PMScreenshot 2024-01-10 at 7.25.33 PM
Screenshot 2024-01-10 at 7.25.14 PMScreenshot 2024-01-10 at 7.25.20 PM

Pipeline

The image processing pipeline is mainly consisted of two parts. The first part is the source image segmentation module, which aims to generate the image mask used to precisely control the modifying part of the image. The second part is the diffusion-based text-to-image model, which will change the part of the image based on the text prompt.  The entire pipeline is shown below. The GitHub repo could be found here.

Result

Next Step

We plan to deploy our tool on Hugging Face, utilizing the platform’s infrastructure to host and share the model. To enhance user accessibility and interaction, we intend to develop a user-friendly website interface using Gradio. We aim to create an intuitive and publicly accessible tool. The technical requirements include configuring the model for Hugging Face deployment, developing the website using Gradio, and managing necessary dependencies for a smooth user experience.

Implication

Through a juxtaposition of tranquil images with AI-processed depictions of disasters like floods, extreme precipitation, fire, drought, and tornadoes, the project seeks to evoke empathy and urgency. Beyond being a disaster visualizer, it holds the potential to aid architects and engineers in risk assessment by leveraging generative AI to simulate destruction scenarios, guiding improvements, prioritizing construction methods, and advocating for the importance of insurance. Furthermore, this interactive tool may contribute to the circular economy through prioritizing sustainable construction methods, including promoting renovation and retrofitting for longevity, emphasize the long-term economic benefits of investing in resilient and sustainable construction.This multifaceted initiative strives to engage a broader audience actively in climate change mitigation, preserving our world’s heritage and natural beauty for future generations.

Leave a Reply