CODEWITHSUNDEEP
pythontensorflow
Quanlisp
Quanlisp

Reputation: 13

How to Let Custom Module Refer to Imported Module on Main Python

Here's my custom module stored in util.py that wrapps up 3-steps what I always gone-through to read images when using tensorflow

#tf_load_image from path 
def load_image(image_path, image_shape, clr_normalize = True ):
    '''returns tensor-array on given image path and image_shape [height, width]'''
    import tensorflow as tf
    image = tf.read_file(image_path)
    image = tf.image.decode_jpeg(image, channels=3)
    if clr_normalize == True:
        image = tf.image.resize_images(image, image_shape) / 127.5 - 1
    else:
        image = tf.image.resize_images(image, image_shape)
    return image

But this is quite inefficient to deal with lots of image load since it generally calls import tensorflow as tf everytime I read "single image".

I'd like to let this function to inherit tf command from the main.py's tf where the load_image actually imported into.

Like..

import tensor flow as tf
from util import load_image

image = load_image("path_string") #where load_image no longer wraps the import tensorflow as tf in itself.

Upvotes: 0

Views: 183

Answers (2)

bruno desthuilliers
bruno desthuilliers

Reputation: 77912

But this is quite inefficient (...) since it generally calls import tensorflow as tf everytime I read "single image".

imported modules are cached in sys.modules on first import (I mean "the very first time a module is imported in a given process"), subsequent calls will retrieve already imported modules from this cache, so the overhead is way less than you might think. But anyway:

I'd like to let this function to inherit tf command from the main.py's tf where the load_image actually imported into.

The only way would be to explicitely pass the module to your function, ie:

# utils.py

def load_image(tf, ....):
    # code here using `tf`

and then

# main.py

import tensorflow as tf
from utils import load_image
im = load_image(tf, ...)

BUT this is actually totally useless: all you have to do is, in utils.py, to move the import statement out of the function:

# utils.py

import tensorflow as tf
def load_image(image_path, image_shape, clr_normalize = True ):
    '''returns tensor-array on given image path and image_shape [height, width]'''
    import tensorflow as tf
    image = tf.read_file(image_path)
    image = tf.image.decode_jpeg(image, channels=3)
    if clr_normalize:
        image = tf.image.resize_images(image, image_shape) / 127.5 - 1
    else:
        image = tf.image.resize_images(image, image_shape)
    return image

FWIW, it IS officially recommanded to put all your imports at the top of the module, before any other definitions (functions, classes, whatever). Importing within a function should only be used as a Q&D last resort fix for circular imports (which are usually a sure sign you have a design issue and should actually be fixed by fixing the design itself).

Upvotes: 1

shamilpython
shamilpython

Reputation: 485

As @MatrixTai has suggested in the comments, you could make utils.py like this:

import tensorflow as tf

def load_image(.........):
    ............

and in main.py, you could import util as you want to.

Upvotes: 0

Related Questions