Memory changed when changing the value of crop_to_aspect_ratio in case that this arg actually not work #21773
Description
Bug Issue
The doc of keras.layers.Resizing() shows its description as below:
keras/keras/src/layers/preprocessing/image_preprocessing/resizing.py
Lines 44 to 50 in 53987a7
For the repro below, when I tried to change crop_to_aspect_ratio from False to True, the memory used turns to be fewer.
I don't know if this is actually expected.
In my opinion, this behavior is not expected for that only when the original aspect ratio differs from the target aspect ratio, the output image will be cropped. The result should be the same anyway.
Repro 1 (crop_to_aspect_ratio==False)
import os
os.environ['CUDA_VISIBLE_DEVICES'] = '3,4'
# tensorflow memory usage test
import tensorflow as tf
gpus = tf.config.experimental.list_physical_devices('GPU')
if gpus:
for gpu in gpus:
tf.config.experimental.set_memory_growth(gpu, True)
# Main Code -->
import tensorflow as tf
import keras
input_image = tf.random.uniform((1, 256, 256, 3), minval=0, maxval=255, dtype=tf.float32)
resized_image = keras.layers.Resizing(height=224, width=224, interpolation='bilinear', crop_to_aspect_ratio=False)(input_image)
print('Resized Image:', resized_image.shape)
# Main Code <--
memory = 0
for i in range(len(gpus)):
memory += tf.config.experimental.get_memory_usage('GPU:%d' % i)
print(memory)Output 1
Resized Image: (1, 224, 224, 3)
2097152
Repro 1 (crop_to_aspect_ratio==True)
import os
os.environ['CUDA_VISIBLE_DEVICES'] = '3,4'
# tensorflow memory usage test
import tensorflow as tf
gpus = tf.config.experimental.list_physical_devices('GPU')
if gpus:
for gpu in gpus:
tf.config.experimental.set_memory_growth(gpu, True)
# Main Code -->
import tensorflow as tf
import keras
input_image = tf.random.uniform((1, 256, 256, 3), minval=0, maxval=255, dtype=tf.float32)
resized_image = keras.layers.Resizing(height=224, width=224, interpolation='bilinear', crop_to_aspect_ratio=True)(input_image)
print('Resized Image:', resized_image.shape)
# Main Code <--
memory = 0
for i in range(len(gpus)):
memory += tf.config.experimental.get_memory_usage('GPU:%d' % i)
print(memory)Output 1
Resized Image: (1, 224, 224, 3)
2096384
Thanks a lot!