The README mentions that the Arizona, New York, and Yosemite cases are free for distribution, but only the Arizona subset (case_in/input_unsorted.csv) appears to be in the repo. Are the NY and Yosemite cases available somewhere else, or could they be added?
Context on why I'm asking. I've been building a terrain-aware probability prior generator for pedestrian SAR (Koester radial plus slope, forest, and linear-feature modifiers, geodesic reachability, and optional directional witness info). Ran it against the Arizona subset and reproduced the 0.78-ish pure-Koester baseline from the 2016 Trans-GIS paper as a sanity check. The Yosemite cases would be particularly useful since mountainous terrain is where the geodesic component should matter most, and Arizona is mostly too flat to exercise it.
Also noticed a few small things worth flagging while I was working with the Arizona data:
- The Distance column in input_unsorted.csv doesn't agree with great-circle distance computed from the IPP and find lat/lon columns. It's systematically smaller by a median factor of 1.53x, ranging from 1.3x to 6x across the dataset. I ended up filtering on haversine from the lat/lon columns instead. Not sure if that's a known issue or whether the CSV distances come from a different source (watershed? some legacy units?).
- case_in/exported_case_Library.txt is empty (0 bytes).
- framework_case in database/website_data.db has 0 rows but a nice schema. Was planning to use it as a reference for what full silver-standard records look like.
Happy to share evaluation code and results back if useful. Thanks for making this available = it's been the one publicly accessible benchmark that let me calibrate against published baselines.
The README mentions that the Arizona, New York, and Yosemite cases are free for distribution, but only the Arizona subset (case_in/input_unsorted.csv) appears to be in the repo. Are the NY and Yosemite cases available somewhere else, or could they be added?
Context on why I'm asking. I've been building a terrain-aware probability prior generator for pedestrian SAR (Koester radial plus slope, forest, and linear-feature modifiers, geodesic reachability, and optional directional witness info). Ran it against the Arizona subset and reproduced the 0.78-ish pure-Koester baseline from the 2016 Trans-GIS paper as a sanity check. The Yosemite cases would be particularly useful since mountainous terrain is where the geodesic component should matter most, and Arizona is mostly too flat to exercise it.
Also noticed a few small things worth flagging while I was working with the Arizona data:
Happy to share evaluation code and results back if useful. Thanks for making this available = it's been the one publicly accessible benchmark that let me calibrate against published baselines.