sbrl-github/research-rainfallradar
1
0
Fork 0
mirror of https://github.com/sbrl/research-rainfallradar synced 2024-12-22 06:05:01 +00:00
Code Issues Projects Releases Packages Wiki Activity

dataset_mono: add water_threshold=None support

Browse source 
This is for the stupid pointless regression thing
Like just let me get on with sample weighting and accounting for extreme event bias already!
This commit is contained in:
Starbeamrainbowlabs 2024-12-12 23:36:54 +00:00
parent 7e00ede747
commit 4dd66a9bdf
Signed by: sbrl
GPG key ID: 1BE5172E637709C2
1 changed files with 4 additions and 2 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

6
aimodel/src/lib/dataset/dataset_mono.py
View file

@ -23,7 +23,7 @@ def parse_item(metadata, output_size=100, input_size="same", water_threshold=0.1
metadata (dict): Metadata about the shapes of the dataset - rainfall radar, water depth data etc. This should be read automaticallyfrom the metadata.json file that's generated by previous pipeline steps that I forget at this time.
output_size (int): The desired output size of the water depth data.
input_size (str or int): The desired input size of the rainfall radar data. If "same", it will be set to the same as the output_size.
water_threshold (float): The threshold to use for binarizing the water depth data.
water_threshold (float|None): The threshold to use for binarizing the water depth data. If None, then no thresholding will be done. IMPORTANT: setting `water_threshold=None` will NOT remove the channels! You gotta do that yourself!
water_bins (int): The number of bins to use for the water depth data (e.g. for one-hot encoding).
heightmap (tf.Tensor): An optional heightmap to include as an additional channel in the rainfall radar data.
rainfall_scale_up (int): A factor to scale up the rainfall radar data.
@ -113,7 +113,9 @@ def parse_item(metadata, output_size=100, input_size="same", water_threshold=0.1
# water = tf.cast(tf.math.greater_equal(water, water_threshold), dtype=tf.int32)
# water = tf.one_hot(water, water_bins, axis=-1, dtype=tf.int32)
# SPARSE [LOSS dice / sparse cross entropy]
water = tf.cast(tf.math.greater_equal(water, water_threshold), dtype=tf.float32)
if water_threshold is not None: # if water_threshold=None, then regression mode
water = tf.cast(tf.math.greater_equal(water, water_threshold), dtype=tf.float32)
# BUG it may be a problem we're [height, width, channel] here rather than [height, width], depending on how dlr works
if do_remove_isolated_pixels:
water = remove_isolated_pixels(water)