Closed Flamefire closed 4 years ago
The following code looks wrong: https://github.com/tensorflow/benchmarks/blob/d0743d98a7f3bb5f88f2fb9cc3fe6b72d4b152d5/scripts/tf_cnn_benchmarks/preprocessing.py#L251-L252
Per definition height / image_height_float == 1 as the conversion to float32 is losless for all meaningfull values
height / image_height_float == 1
I guess something like image_width_float / image_height_float was intended
image_width_float / image_height_float
Oh, my bad, sorry. Got confused about all the values
The following code looks wrong: https://github.com/tensorflow/benchmarks/blob/d0743d98a7f3bb5f88f2fb9cc3fe6b72d4b152d5/scripts/tf_cnn_benchmarks/preprocessing.py#L251-L252
Per definition
height / image_height_float == 1
as the conversion to float32 is losless for all meaningfull valuesI guess something like
image_width_float / image_height_float
was intended