GeoAI Amazon Basin Secret Runway Detection Challenge by ITU
and 1 other country
- Austria
- Peru
- Italy
- Scroll to see more
Hi, Thanks for organizing this fun competition!
My understanding of the scoring in this competition is that we will submit a csv with the first column ("tile_row_column") being identical to SampleSubmission.csv and the second one ("label") containing our prediction for each tile.
And so, one AOI, eg 23_01 contains a bit over 1000 * 1000 tiles, arranged in approx. 1000 columns and 1000 rows. My questions, specifically are the following:
- How many rows and cols are there in one AOI?
- From which corner are they counted, ie is Tileaoi_23_01_1_1 in the bottom-left corner of AOI 23_01, or the top-left one, or the top-right one, or the bottom-right one?
- Does every tile in one AOI have the same area on the globe, or does each one have the same area on a Mercator projection?
Maybe I missed it, but I was unable to find this info anywhere.
Best regards, Emil
You're right about the limited information. I have noticed that all the images are the same size - 1527 * 1541, which also aligns with the numbering in the sample submission file. The CRS projection is in metres, so the satellite image resolution should be 10m * 10m, and each tile is actually a pixel in a 1527 * 1541 image.
Regarding the start and end point, I'm not sure how that works yet, guess I'll figure it out when I make a submission.
This is my understanding; hope we get some answers from the organizers.
@da_ If i may ask.
1. How did u generate Grid Cells for each AOI. My test data prediction part keep on crashing my colab notebook. I don't know if it's as a result of this. These are what i used generate this grid cells:
## Generate 1527 * 1541 Grid Cells for each AOI with 10m resolution
cell_size = 10 # Fixed cell size of 10m
num_rows = 1527
num_cols = 1541 respectively. Even after i reduce samples for final output sub file from 200,000 to just 100 samples, my notebook still crashed.
2. Moreover, even when i created a customised function to calculate num_rows and num_cols dynamically from test data. Still crashed....
For example:
def calculate_grid_size(minx, miny, maxx, maxy, cell_size):
I downloaded the AOIs as images at 10m resolution, so they come predefined as 1527 * 1541 when I run my download script. I still haven't figured out the image's origin for the submission file.
Thanks a lot da_!
I've noticed one of the areas actually seems to be a bit smaller than the other ones. IDK if it'll be of practical relevance. My calculations gave me this: (NB the last one)
Shapefile Area (sq m) Width (m) Height (m)
aoi_2021_04.shp 235310700.0 15410.0 15270.0 1 aoi_2022_01.shp 235310700.0 15410.0 15270.0 2 aoi_2024_01.shp 235310700.0 15410.0 15270.0 3 aoi_2020_02.shp 235310700.0 15410.0 15270.0 4 aoi_2021_01.shp 235002500.0 15410.0 15250.0 5 aoi_2020_03.shp 235310700.0 15410.0 15270.0 6 aoi_2020_01.shp 235310700.0 15410.0 15270.0 7 aoi_2022_02.shp 235310700.0 15410.0 15270.0 8 aoi_2023_01.shp 235310700.0 15410.0 15270.0 9 aoi_2021_03.shp 235310700.0 15410.0 15270.0 10 aoi_2021_02.shp 235002500.0 15410.0 15250.0 11
After investigating a bit more i found that in sample submission there are
- 1541 different values for rows
- 1527 different values for columns
However, the AOIs have dimensions
- 15270 meters height
- 15410 meters width
Was there maybe a mixup between columns and rows in the csv?
...because this would mean that a tile has dimensions 10.09m x 9.91m instead of 10m x 10m.
I think you are right about the mixup in column and rows in the csv file. We need confirmation from the organizers.
@meganomaly Could you kindly clarify the submission file nomenclature? Are the rows and columns flipped as observed. Also, what is the origin for indexing, top_left or bottom_left? Please, and thank you.