Add SpaceNet5 #263
Add SpaceNet5 #263
Conversation
|
|
||
| if self.speed_mask: | ||
| val = speed_cls_arr[ | ||
| int(feature["properties"]["inferred_speed_mph"]) - 1 |
There was a problem hiding this comment.
speed_cls_arr is an array of length 65 with a class corresponding to each index/speed mph. Essentially its a mapping between speed mph (index) and the class. -1 is to account for index starting at 0.
| """ | ||
| min_speed_bin = 1 | ||
| max_speed_bin = 65 | ||
| speed_arr_bin = np.arange(min_speed_bin, max_speed_bin + 1) |
There was a problem hiding this comment.
np.arange(min, max) produces an array ranging from min to max-1. +1 is to include the max
| speed_arr_bin = np.arange(min_speed_bin, max_speed_bin + 1) | ||
| bin_size_mph = 10.0 | ||
| speed_cls_arr = np.array( | ||
| [int(math.ceil(s / bin_size_mph)) for s in speed_arr_bin] |
There was a problem hiding this comment.
Why are we binning these? Wouldn't it be better to leave them as floats and use it in a regression problem?
There was a problem hiding this comment.
Binning is done to create a multi channel speed mask which would enable training a segmentation model. This was the strategy adopted by the SpaceNet authors (cresi) and the winning solutions hence why I opted for this.
|
Good to merge @ashnair1? |
@calebrob6 Good to go 👍 |
adamjstewart
added
utilities
* Add SpaceNet5 * Add speed_mask option and test data * Fix docs * Combine fixtures * Use python copy * Remove list * Predictions -> Labels * Better error message
_load_maskandplotoverrided to support multi channel speed masks.Sample Plots:
Binary Mask (With speed_mask=
False)Speed Mask (With speed_mask=
True)By default, every 10 mph constitutes a class hence 7 classes (not including background).