How to get the geo coordinates of a pixel within a bounding box?

[rsousa95]

Sorry if this is a beginner question, but I'm not finding the answer anywhere.

I created a RasterDataset and I query it the section I want via a BoundingBox. It returns me a dictionary containing:
image, a torch tensor with the image;
crs, the coordinate reference system;
bbox, the bounding box information regarding the section in question.

Now, I need to get the geographic coordinates of a pixel point within this image. Let's say I want the central point. I'd want something like
center_x, center_y = get_coordinates_of_pixel(x=image.width//2, y=image.height//2, crs, bbox).
How can I do this?

[adamjstewart]

You can get this information from the bounding box:

bbox = sample["bbox"]
center_x = (bbox.minx + bbox.maxx) / 2
center_y = (bbox.miny + bbox.maxy) / 2

[calebrob6]

The offsets for each pixel can be calculated in terms of the RasterDataset's res property, i.e. you should be able to do something like bbox.miny + (y * ds.res).

[adamjstewart]

Ah yes, my answer is for geographic coordinates, @calebrob6's answer is for pixel coordinates. Wasn't sure which you were asking about.

[adamjstewart]

Nvm, both of these are for geographic coords and should give the same answer.

[rsousa95]

Both of these got me what I wanted, thank you! @calebrob6's answer is the more generic one for any point but both work.

Just to add something in case someone stumbles upon this in the future, I did have to do something like:
geo_x = bbox.minx + (pixel_x * dataset.res)
geo_y = bbox.maxy - (pixel_y * dataset.res)

This is because the reference systems are different. From what I understand, image tensors have their origin in the upper left corner, x axis to the right, y axis downwards. The tiff I was working on has the origin in the lower left corner, x axis to the right, y axis upwards. Hence this translation and inversion in the y axis.