Digital Resurrection: Restoring Historical Buildings using Artificial Intelligence
Advisor: Meyerowitz
Internship Location: PAC-SJ "Endangered Eight" buildings(my advisor runs CP Mentorship)
Onsite Mentor: Catherine Liu, CP Mentorship
Throughout the world, historical buildings are fading away due to factors such as gentrification, natural disasters, and war. This puts the rich history that they contain in jeopardy. Additionally, it is very expensive and time consuming to manually preserve the building, leading to debates over whether or not the preservation of historical buildings is even necessary. In recent years, some have posited the idea of digital restoration in order to more efficiently preserve these buildings, but I believe this process can be further streamlined. I aim to answer if the digital restoration historical buildings using cutting-edge technologies such as virtual 3D modeling and artificial intelligence (AI) is possible.
My only experience with historical preservation comes from my work with the community of Alviso in San Jose. Since the 10th grade, I have been working with my onsite advisor and members of the Alviso community to preserve their history, mainly through a documentary about Thomas Foon Chew, who was a very important figure in the community's history. This led me to his old Bayside Canning Company building, a historically significant building that is very out of shape and will most likely be damaged beyond repair within the next 5 years. This project was born out of a desire to save that building. I have already made a prototype of the building, but it is very un-detailed, basic, and unpolished. Thats why I want to use AI to streamline and improve the model.
I will start off with some library and internet research to better understand the technology I will be working with. I will then begin prototyping some models to test said technology out. My goal is to have the "endangered eight" all digitally modeled and put online by the end of the three months. I will break down my research into 12 weeks. Weeks 1 and 2 will be used to understand the technology I need to 3d model buildings with AI. Week 3 will be used to do a test model of my house. Week 4 will be used to troubleshoot and improve on the model. From then on, every week will be dedicated to 3d modeling one of the "endangered eight" buildings, as I should have all the tools I need to streamline the process.
Project Blog
Week 1: New Horizons
Context
For a long time, history was just a hobby to me. Something that was experienced through Wikipedia and trivia competitions. It wasn't until January of 2024 when that changed.
I made a wrong turn into Alviso. Once a thriving port that ferried the South Bay’s harvests to San Francisco, Alviso now felt like an overlooked corner. Unpaved, pothole-filled roads led past abandoned warehouses, and the nearest grocery store was miles away. The Alviso I read about in the books and the Alviso my eyes were staring at could not have been more different.
Curious, I returned, notebook in hand, chasing answers that only multiplied: floods, lawsuits, and decades of neglect. One day, Francisco, a lifelong resident, pointed to a derelict factory building and joked, “This was our Google 100 years ago, but now it’s our museum.” His words struck me. For the first time, history was rusting right in front of me, and I had to save it. That impulse soon blossomed into a website and my research paper for National History Day.
My experiences in Alviso showed me that history is dying. Throughout the world, historical buildings are fading away due to factors such as gentrification, natural disasters, and war. This puts the rich history that they contain in jeopardy. Additionally, it is very expensive and time consuming to manually preserve the building, leading to debates over whether or not the preservation of historical buildings is even necessary. In recent years, some have posited the idea of digital restoration in order to more efficiently preserve these buildings, but I believe this process can be further streamlined. I aim to answer if the digital restoration historical buildings using cutting-edge technologies such as virtual 3D modeling and artificial intelligence (AI) is possible.
What I did this week
This week wasn't all that eventful(senior projects are just beginning), but I was still able to get some preliminary work done. For starters, I contacted the owners of the "Endangered Eight" properties and got some responses. I'll be spending the next two weeks trying to get all the green lights I need for my onsite work. In addition, I've also been learning to use the three softwares my project will revolve around(Artlist, Photoshop, and Blender). Honestly, I think my skills right now are good enough, but I do want to make sure they're polished enough while I'm still on a relatively loose schedule.
I guess I also met with my internal and external advisors, but that's a given every single week. In future blogs, I won't explicitly mention this unless I wasn't able to meet with them.
Research Question
My original research question was "How can certain AI models such as generative AI and image to image be utilized to more efficiently model and digitally preserve historical buildings?" However, during my proposal, someone suggested that my project didn't involve that much artificial intelligence to begin with.
Should I roll with this question and if not, how should I rephrase it?
Goals for next week
Finish learning to use Artlist, Photoshop, and Blender
Read Multiple View Geometry in Computer Vision and Historic Preservation: Curatorial Management of the Built World.
Week 2: What exactly is Digital Resurrection?
Goals for last week
Finish learning to use Artlist, Photoshop, and Blender - I'm actually getting pretty good at these tools, and I'll make sure to demonstrate me using these in action once I add the finishing touches on my methodology
Read Multiple View Geometry in Computer Vision and Historic Preservation: Curatorial Management of the Built World - I was able to get a lot of extra reading done, which is why I decided to make this blog post a full literature review for my project.
Overall, I believe I met and surpassed my goals from last week. Obviously, not every week is going to be like this, but my project is going quite swimmingly if I do say so myself.
Literature Review
This is, for the most part, all the reading I will be doing for my project, unless something interesting catches my eye in the upcoming weeks.
You'll also notice how I don't completely go into detail for every single one of my sources. This is because a lot of the more theoretical sources only provide hypothetical methodologies that artificial intelligence improves on in every way possible. I know I'm sounding very vague and cocky, but I'll go more into detail when I demonstrate my methodology in my next blog.
Lahaina
The article explains how architecture students at the University of Hawaii are using artificial intelligence to reconstruct the town of Lahaina, that was lost in a wildfire in 2023. This is, by far, the closest project I have seen to what I aim to do. However, their methodology differs from mine in two ways. First, they use completely different softwares from me. From what I've seen, their primary 3d modeling software is Lumion, while mine is Blender. This also leads into the second difference, which is that they are not generating their models from images. It'll be interesting to see if my approach of forming my reconstructions with the use of image-to-image AI will create models more detailed than theirs, but their work is already so detailed that I'm starting to doubt my methodology a bit. It's quite a hard bar to clear, as you can see from their model below.
Notre Dame
This article was similar to the Hawaii one, but the methodology is different. For context, this article is about efforts to digitally reconstruct the Notre Dame Cathedral in Paris following the 2019 fire that burned it down. Several teams worked to reconstruct it, but the most prevalent one seems to have used the software Solidworks. Again, I am using different softwares and adding an image-to-image component, which should enable me to make more accurate and detailed models of even older buildings.
Multiple View Geometry in Computer Vision and Historic Preservation: Curatorial Management of the Built World
The reason I put these two works together is that they are very closely related in the sense that they are both theoretical works about this field that were published years before artificial intelligence became so mainstream. Due to this, most of their content is relatively outdated, as many of the problems they posit can be solved through any combination of the three softwares I am using. For instance, Multiple View Geometry centers around a complex plan for turning 2d images into 3d models, while my methodology can eliminate several steps in said plan.
However, the ethical questions they bring up are interesting and very much worth discussing. As a result, I will now be dedicating a section in all my future blogs just rambling about philosophical questions related to historical preservation. I'll spare you all for this blog, as it's getting quite long.
Citations for Literature Review
“Students Use AI to Preserve Lahaina’s Historical Architecture.” University of Hawaiʻi System News, 22 Mar. 2024, www.hawaii.edu/news/2024/03/22/lahaina-historical-architecture/.
Tyron, Leon. From Ruins to Residences: How Ai Is Revolutionizing the Refurbishment of Old Buildings into Modern Homes | by Leon Tyron | Medium, medium.com/@leontyron/from-ruins-to-residences-how-ai-is-revolutionizing-the-refurbishment-of-old-buildings-into-modern-a46d0b3e32d3. Accessed 11 Jan. 2026.
Reconstruction of Historical Cities Using AI Technology (Augmented Reality), www.researchgate.net/publication/386354884_Reconstruction_of_Historical_Cities_Using_AI_Technology_Augmented_Reality. Accessed 15 Jan. 2026.
Hartley, Richard, and Andrew Zisserman. Multiple View Geometry in Computer Vision. Cambridge University Press, 2019.
Fitch, James Marston. Historic Preservation: Curatorial Management of the Built World. University Press of Virginia, 2007.
Horgan, Rob. “How Digital Technology Is Aiding Notre Dame Rebuild, Two Years since Tragic Blaze.” New Civil Engineer, 15 Apr. 2021, www.newcivilengineer.com/latest/how-digital-technology-is-aiding-notre-dame-rebuild-two-years-since-tragic-blaze-15-04-2021/.
Goals for next week
Finalize methodology enough to make a prototype model
Get access to Alviso's Bayside Canning Company(the first building I aim to model)
Week 3: 3 steps to resurrect history
Goals for last week
I cheated a little this week. By that, I mean that I went ahead of schedule while not fully finishing the goals I set for last week. Let me explain. My goal was to finalize my methodology and gain access to Alviso's Bayside Canning Company. My methodology is not finalized yet, but not only have I gained access to the Canning Company, I've also made a preliminary model.
Nevertheless, I am still on track for my big schedule, as my goal by the end of week 4 was to have my methodology and first model completed. I just went about it in a different order than planned.
Methodology
I have decided to combine my methodology and model display into one section. More specifically, I will be going through my current methodology while demonstrating how my preliminary model for the Bayside Canning Company took shape.
Artlist
The first software in my procedure is Artlist, an image to image AI model. Essentially, it allows me to combine and modify images in whatever way I choose.
Here's Artlist in action with my model. Below, there are two pictures. The picture on the left is the unmodified and current version of the the Canning Company, while the picture on the right combines the current day image with an older image to generate a more accurate representation of what the company would've looked like in the 1920s.
Blender
To actually create the final model, I have decided to use the 3d modeling software Blender. I chose it because of it's unique ability to generate a 3d model from 2d images, which allows it to pair perfectly with Artlist. To summarize, Artlist modifies the images to reconstruct the "prime" version of all of the building's walls, and Blender merges the walls and other details together to create the model.
Here is my current Blender model of the Canning Company. Over the course of next week, I'll try to make it more complete.
What's missing
I am still trying to figure out a model that can bridge Artlist and Blender together. What I mean is that although Artlist's image modifying abilities are incredibly effective, they can leave out some of the finer details on a building. I want to make my reconstructions as accurate as possible, so I will need another software to modify very specific details each image. The software I have in mind is Adobe Photoshop, but I am experimenting with other softwares too. Hopefully by the next blog, I'll be able to show my completed procedure.
A discussion on ethics
As promised last week, I am dedicating a space in all my future blogs to discussing the ethics of what I'm doing. Specifically, I wanted to talk about the rapid improvement in AI models. In all of the source materials I've read, the central question they posit is how to accurately reconstruct models from the past. Now, image to image models make this issue far more manageable. But this begs the question: If AI models continue to improve at such a rapid pace, will there even be a need for humans who specialize in digital historical reservation?
To answer this question, I'll break out my inner linguist. History is a humanities subject. Notice how you can't spell humanities with the word "human." Ok, that was pretty lame, but my point is that history is a subject that has exclusively been experienced by humans. If we entrusted the field of historical preservation to AI, then there will be no one left to fact check or edit the finer details that AI could miss, which could lead to important historical information being lost forever simply because AI overlooked or deemed it "not important."
Basically, no matter how powerful AI becomes in the future, there will always be an important role for humans to fill in the field of historical preservation.
Goals for next week
Finish methodology(specifically figure out how to integrate an image editing software)
Finish model of Bayside Canning Company
Week 4: The Bayside Canning Company
Goals for last week
We are officially back on track. Last week, I integrated Photoshop into my methodology and was able to complete my model of the Bayside Canning Company, an endangered building in San Jose's Alviso neighborhood.
Methodology
Just a recap of my methodology, now with Photoshop. To spare some time, I won't be going into depth about Artlist and Blender. If you are curious about those softwares, check out my blog from last week.
Artlist
Artlist is an image to image AI model that allows me to combine images together. I use it to create images of a building's past state by combining its older photo with its more detailed newer photo, which gives the perfect balance of quality and historical accuracy. The left photo below is the Bayside Canning Company's current state, while the right photo is the "Artlisted" version.
Photoshop
Photoshop is an image editing software that will serve as the bridge between Artlist and Blender. The central problem with Artlist is that it leaves out some of the finer details on its created images. Photoshop will allow me to add those finer details and fix Artlist's potential mistakes.
Blender
Blender is a 3d modeling software that can generate a 3d model from 2d images, which allows me easily and efficiently create a model from my modified photos.
Below is the prototype Blender model of the Canning Company.
The model
Here are various angles of my finished model. My end product will be a digital exhibit with all my models, but that's for a later blog.
A discussion on ethics
Honestly, there's not much for me to talk about here, so I'll instead dedicate this space to talking about the history of the Bayside Canning Company and its founder Thomas Foon Chew.
Born in 1889 in China, Foon Chew emigrated to the United States when he was very young. He overcame the harsh anti-Chinese sentiment of the Bay Area at the time to found the Bayside Canning Company, eventually turning it into the third largest cannery in the United States.
However, Foon Chew was better known as a philanthropist. His kindness and honesty impacted pretty much the entire Bay Area. He famously provided Alviso with free water as long as he was alive, and was deeply involved in San Francisco's Chinatown. The most telling sign of his reputation was that, upon his death, 25,000 people, including his biggest competitors, attended his funeral.
I personally believe Foon Chew deserves a place in our history textbooks.
Goals for next week
Begin model of the South Bay Yacht Club
Week 5: Slowing things down
Goals for this week
My main goal for last week was to start on my model of the South Bay Yacht Club, and I was able to get some work done. Most notably, I'll be able to spend this week using drones to map it out. However, I want to slow things down going forward, as I believe that doing some more research to back up my methodology will ultimately lead to better results.
Methodology(framework)
Both my advisors have advised me(so creative I know) to connect the dots between my literature review and my methodology, and looking at my past blogs, I've realized that said connection is my number one priority. Hopefully, my endless methodology saga can finally end this week.
In "Multiple View Geometry in Computer Vision," which is my primary theoretical source regarding the use of computers in historical reconstruction, the authors lay out the limiting factors of complete digital reconstruction. First, it was extremely complicated and time consuming to turn 2d images into 3d models using the technology of their time(for reference, the book was published in 2004), as it required the implementation of complex algorithms fairly unaccessible to the general public. Second, there was simply no way to make high quality models of buildings in their past state, as simply using a building's old picture would lead to very sloppy models.
As the years progressed, many researchers began coming up with solutions to these two main problems. For starters, new and improved softwares such as Lumion and Blender were created with the specific purpose of creating 3d models out of 2d images. As these softwares improved, they became much more accessible to the general public, which essentially solved the first problem. However, the second problem was far more tricky. Teams that reconstructed the Hawaiian town of Lahaina after the 2023 wildfires had access to fairly recent photos of the town before it was destroyed, allowing them to simply use those pictures to create the model. As for those in charge of digitally reconstructing the Notre Dame Cathedral, they were able to end up with high quality models due to utilizing portions of already existing digital models of the cathedral, such as the one found in the video game series Assassins Creed.
Where I come into this equation is my use of Artlist, an AI model specifically made to combine images together. This will allow me to use far older images and make up for their lack of detail by combining them with newer images. My use of Photoshop also serves as a bridge between Artlist and Blender, the software that I will be 3d modeling with.
The South Bay Yacht Club
Again, I am using this section about "ethics" to dump some historical knowledge. The South Bay Yacht Club itself is a very interesting building to model, as it was originally built in 1903 before floods forced its members to move it up a hill where it stands today. As a result, I will not only have to deal with the building, but its surroundings as well. Whether or not I can accurately depict the clubhouse's original surroundings is a question I am very eager to answer
Goals for next week
Map out South Bay Yacht Club
Week 6: A case study in floods
Goals for last week
First of all, I'd like to apologize for my extended leave of absence. It won't happen again, I swear.
My main goal from the last blog was to start on my model of the South Bay Yacht Club, and I accomplished that. I have been granted permission by it's owners to fly drones on the property. Stay tuned for that
A comparison of the buildings
Since there's nothing really interesting about taking pictures of buildings, I'll use this blog to analyze the impact of floods on historical buildings. It sounds niche, but natural disasters are one of the leading causes of damaged historical buildings, and both of the buildings I've worked on so far have suffered from it.
First, it's important to provide some context. Both buildings are located in a neighborhood called Alviso, which happens to be the lowest point in the Bay Area that sits at around 10-15 feet below sea level. This makes it extra prone to floods, with major ones in in 1983 and 1995 dealing severe damage to the small community.
But if you went to Alviso, you'd notice that the South Bay Yacht Club appears in great shape while the Bayside Canning Company is, to put in nicely, far removed from its glory days. So, what happened? Well, as I mentioned in my last blog, the South Bay Yacht Club was rebuilt and moved up a hill following a major flood. With this in mind, it is very easy to see how the Yacht Club has remained in a better condition than the Canning Company.
So, what's the big lesson? Well, it's that historical preservation is a very delicate field that is impacted by a variety of factors. Even a couple feet of difference in altitude can have dramatic effects on a historical building's condition. Hopefully, the ever evolving field of digital preservation will be able to at least somewhat curb these factors, because boy there are a lot of them.
Goals for next week
Use Artlist on all sides of the South Bay Yacht Club
Week 7: Problems with the Yacht Club
Goals for last week
Last week, I shifted my focus to Alviso’s South Bay Yacht Club. While I haven’t started modeling yet, I was able to get all the images I needed from San Jose’s King library and the club itself using drones.
The Challenge of Location
Unlike the Bayside Canning Company, the South Bay Yacht Club is a step up in difficulty due to its location
The building sits in Alviso’s tidal marshlands, right on the edge of the San Francisco Bay. This creates several complications:
Environmental Wear and Change
The surrounding landscape is constantly shifting due to tides, erosion, and sediment buildup.Limited Angles
Because the yacht club is relatively isolated, there are fewer historical photographs taken from diverse perspectives. Most images come from similar vantage points, which means I can only get a high quality model of a couple sides.Lighting and Distortion Issues
Many photos include glare from water, fog, or haze—common in the Bay—which can obscure architectural details and introduce inconsistencies when using AI tools like Artlist.
In short, this model will be the perfect test to see how my methodology holds up under less-than-ideal conditions.
Methodology (Applied to This Project)
Speaking of methodology, let's talk about it.
Artlist will need more careful prompting and comparison due to environmental inconsistencies
Photoshop will play a larger role in reconstructing missing or obscured details
Blender will involve a lot more manual work when translating flat marshland imagery into a 3D space
In short, this model will require less automated tasks and a more hands on approach. I'll need to get more involved in every single step if I want it to turn out well.
Goals for Next Week
Complete the 3D model of the South Bay Yacht Club
Week 8: The end?
Goals for last week
Unfortunately, I was not able to complete my model of the South Bay Yacht Club, and the "model" I have right now is not nearly ready for presentation. Yeah, not a great start to this blog.
Reflection
I honestly don't have much to talk about in this blog, so I guess I'll use it to reflect on what I learned these past few weeks.
First, I overestimated myself greatly, I believed that I could easily model 8 buildings, but was only able to get one done. There are many reasons for why this happened, but I'm not here to make excuses. I should have done more but didn't, and I'm pretty disappointed in myself for that.
But let's not dwell on the negatives. I did come out of this project having gained a new appreciation for historical preservation, knowledge on how to use several helpful tools, and a desire to expand on this project in college. It would be a lie to myself if I said the past few weeks were purely a disappointment. I genuinely enjoyed doing this project, and want to use the methodology I've created to model my planned buildings over the summer and into my future.
Overall, the experience I had was great and I'm content with the end product I put out, but I'm definitely bummed about not being able to accomplish more in this timeframe.
End Product
So what is this end product I'm talking about. Well, there's two components to it. The first is the methodology I've created involving Artlist, Photoshop, and Blender. It worked very well for the Bayside Canning Company, and I firmly believe a little tuning is all it needs to be able to model more advanced buildings.
The second component is, of course, the Bayside Canning Company model itself. Right now, I'm working on a "digital exhibit" to put it in by the time of my big presentation, but you can look at my past blogs to preview it.
That's all from me, I guess. These past 12 weeks have been a blast, and in case I don't see you, good afternoon, good evening, and good night.
Goals for Next Week
Deliver a fantastic presentation
Week 9: Future Research
Some additional thoughts
After giving my presentation, I realized I had a lot of thoughts on the direction the field of digital historical preservation could go towards in the future. Here they are.
Structural Predictor AI: Maybe we could use AI to scan a historical building and give a timeline as to when we could expect it to undergo some damage. I believe that this would make historical preservation much more efficient and less time-consuming, making it more appealing for others to devote their time towards.
AI with VR: We could also turn these historical models into VR simulations for more people to interact with, making historical sites accessible to those who cannot travel to see them in person.
These are just what I'm thinking about right now, but more can definitely be done in the future. If these innovations came out in the next 20 years, I'd be very impressed.
Week 10: Why should we care
Final Thoughts(for real this time)
At the beginning of this project, I put a little section in my first blog about why people should care about historical preservation, and I just wanted to expand on it a little bit. If anything, I'd like for the main takeaway of my project to not necessarily be how AI can be used to reconstruct historical buildings, but rather that historical preservation and history itself are important.
Historical preservation plays a vital role in protecting the cultural identity and collective memory of societies around the world. Historic buildings, monuments, and neighborhoods serve as physical connections to the past, allowing future generations to better understand the people, traditions, and events that shaped their communities. In many countries, preservation is also deeply connected to tourism, which can become a major source of economic activity. Cities and nations with well-preserved historic districts often attract millions of visitors seeking authentic cultural experiences, generating revenue for local businesses, museums, restaurants, and preservation initiatives themselves. Locations such as ancient temples, colonial architecture, historic ports, and industrial landmarks help transform history into something tangible and accessible rather than something confined to textbooks.
Beyond economics, historical preservation can also encourage cultural rediscovery and renewed appreciation for overlooked histories. Restoring endangered structures or digitally reconstructing lost sites allows communities to reconnect with traditions, architectural styles, and local stories that may otherwise disappear over time. This process can strengthen cultural pride and provide educational opportunities for younger generations who may have little direct exposure to their region’s heritage. In places affected by modernization, natural disasters, or conflict, preservation efforts can help recover pieces of cultural identity that were damaged or forgotten. As technologies such as artificial intelligence and digital modeling continue to evolve, they create new opportunities to document and share historical environments with broader audiences, helping preserve not only buildings themselves but also the cultural narratives attached to them.
Anyways, that's all I got. For real this time, I'm signing off.
Post 11: Post Presentation: What I learned
Alright, let's talk about what exactly what I got out of this project.
Over the course of this three-month project, I learned that preserving historical buildings is about much more than simply recreating old structures. At first, I saw artificial intelligence mainly as a technical tool that could help speed up digital modeling and reconstruction. As I worked with historical photographs, architectural references, and 3D modeling software, though, I began to understand how much history and meaning are attached to these places. Every building has its own story, and trying to recreate missing details forced me to think carefully about the people who once used these spaces and the role they played in their communities. AI helped make parts of the process more efficient, but I also realized that preserving history still depends heavily on human judgment, research, and interpretation.
One of the biggest things I learned was how difficult preservation can be when information is incomplete or damaged. Many historical buildings do not have detailed blueprints or extensive photo archives, so reconstructing them often required piecing together clues from multiple sources. There were times when I had to compare old images, study architectural patterns, and make educated guesses about certain features. That process gave me a much greater appreciation for the work historians and preservationists do every day. It also showed me how quickly important landmarks can disappear because of neglect, urban development, or environmental damage. Working on endangered buildings made the project feel more meaningful, since it highlighted how fragile cultural history can be if efforts are not made to preserve it.
Most importantly, this project helped me understand why historical preservation matters in the first place. Like I've said time and time again, the biggest problem the field of historical preservation is that no one cares. Preserving buildings is not just about saving physical structures; it is about protecting the stories, traditions, and identities connected to them. Historic sites can help communities stay connected to their past while also teaching future generations about local culture and history. I also learned how technology can make preservation more accessible by allowing people to experience places digitally that they may never be able to visit in person. In many ways, this project changed how I look at old buildings. Instead of seeing them as outdated structures, I now see them as important pieces of cultural memory that deserve to be protected and remembered.