bao18 / open_earth_map Goto Github PK

Hi, Bruno

Could you provide a list about the origanization of the dataset? I mean I want to know where does the each images come from.
Could you release a list to index each images to their origin dataset?

There is a interesting question about the UDA task. Do you think the dataset gap(like different satellite) is a domain problem? Will it make a more realistic UDA task for a real application?

I downloaded the dataset from the Zenodo link. It seems like only the Training and Validation splits are included.

Is it possible the request the test set labels ? Thank you!

https://github.com/bruno1866/open_earth_map/blob/f16da336f8abf549e94a206ef67ab869654a557e/data/compile_xbd.py#L41

this line should be == 2 instead of != 2

Hello, I have a question about submission. Should we put all the pre-edited images in the validation data in a .zip file? THANKS

Hello, @bao18 . This dataset is great. But when I use U-Net+efficientb4, I can only obtain 60% IoU, which is far from the 68% as reported in your paper. Do you know why it happens?
Settings: the training parameters are the same as demo_training.py and the only difference is the model.

Colab's network is not friendly to me, so can you make python(.py) on computer?
Can you please add the code how to output the result?
thanks

Hello,

I wanted to ask about the projection information for the labels.

The label GeoTIFFs contain information with the EPSG:4326 CRS (WGS 84). When I render these with respect to other image datasets, I am seeing a misalignment. I tried playing around with linear displacements and it appears to require a more complex reprojection to account for the misalignment.

Based on Arxiv pre-print, I believe this is probably the result of precisely aligning the label with the reference imagery. Is that the case, or do I just have the projection data messed up?

Thank you! This is an incredible dataset!

Hey I am still having issues setting up the dataset correctly. When I try to execute the script to compile the images from xDB I am just getting thrown an error:

(openearthmap) D:\open_earth_map-main>python data/compile_xbd.py --path_to_OpenEarthMap "D:/OpenEarthMap_wo_xBD" --path_to_xBD "D:/xview2_geotiff/xview2_geotiff"

Traceback (most recent call last):
File "D:\open_earth_map-main\data\compile_xbd.py", line 36, in
idx = np.where(xbd_fullimgs[:, 0] == f1)[0]
IndexError: too many indices for array: array is 1-dimensional, but 2 were indexed

The readme says:
The "xbd_files.csv" contains information about how to prepare the xBD RGB images and add them to the corresponding folders.

But I see nothing there but pairs. No further explanation.

@bao18 Can you help me please?

After I carefully prepared all the requested data, I found label file for xBD images missing. The amount of label file in downloaded "OpenEarthMap_wo_xBD" folder is 3500. Where can I get them?