3 ms·

Has anyone worked on AI/ML approaches for detecting ancient structures in LiDAR data? Given that training data for identifying remnants of man-made structures (

by folli 2y ago

Has anyone worked on AI/ML approaches for detecting ancient structures in LiDAR data? Given that training data for identifying remnants of man-made structures (e.g., Roman or medieval ruins) is quite sparse, how would you approach this problem? Some initial thoughts: - Data Augmentation: Using synthetic or simulated LiDAR data from known structures to improve training. - Few-Shot Learning / Transfer Learning: Training models on better-documented archaeological sites and applying them to new areas. Would love to hear thoughts from people with experience in remote sensing, computer vision, or archaeology!

archaeoscape 2y ago
We work on exactly that, in application to ancient Khmer civilization (9th to 15th century). In fact, data is not that sparse, but it's hard to get it. Archaeologists just don't share data as much. The basic answer is that you have to have at least some data to be able to do anything. We found that in the data regime augmentation works a little bit, transfer learning much less so. What really works is simply sitting down and annotating the data, with on-the-ground surveys and follows ups.
- folli 2y ago
  Anything you can share? Papers? Examples?
  - archaeoscape 2y ago
    Sure, you can check our paper where we describe the data scarcity issue, propose a dataset and some baselines: https://arxiv.org/abs/2412.05203 https://arxiv.org/abs/2412.05203 Here's also an interesting LiDAR segmentation paper for Maya settlements. Smaller scale, but you can see what people are trying: https://arxiv.org/abs/2208.03163 https://arxiv.org/abs/2208.03163 Here's another curious recent LiDAR paper, no computer vision, but shows the importance of the proper data processsing: Auld-Thomas L, Canuto MA, Morlet AV, et al. Running out of empty space: environmental lidar and the crowded ancient landscape of Campeche, Mexico. Antiquity. 2024;98(401):1340-1358. doi:10.15184/aqy.2024.148

by folli 2y ago

archaeoscape 2y ago

We work on exactly that, in application to ancient Khmer civilization (9th to 15th century). In fact, data is not that sparse, but it's hard to get it. Archaeologists just don't share data as much. The basic answer is that you have to have at least some data to be able to do anything. We found that in the data regime augmentation works a little bit, transfer learning much less so. What really works is simply sitting down and annotating the data, with on-the-ground surveys and follows ups.

folli 2y ago
Anything you can share? Papers? Examples?
- archaeoscape 2y ago
  Sure, you can check our paper where we describe the data scarcity issue, propose a dataset and some baselines: https://arxiv.org/abs/2412.05203 https://arxiv.org/abs/2412.05203 Here's also an interesting LiDAR segmentation paper for Maya settlements. Smaller scale, but you can see what people are trying: https://arxiv.org/abs/2208.03163 https://arxiv.org/abs/2208.03163 Here's another curious recent LiDAR paper, no computer vision, but shows the importance of the proper data processsing: Auld-Thomas L, Canuto MA, Morlet AV, et al. Running out of empty space: environmental lidar and the crowded ancient landscape of Campeche, Mexico. Antiquity. 2024;98(401):1340-1358. doi:10.15184/aqy.2024.148