Preprocessing a image for yolo pretrained net

Question

I am currently working on a project where I need to use a yolo3_mobilenet1.0_coco but the input image needs to be preprocessed from a format (300,300,3) to (1,3,300,300) which is not being possible by im.resize((1,3,300,300)) but there are particular functions in gluoncv class as data.transforms.presets.ssd.load_test("string_containing_image_file name", short=512) but they all take "string of image file name" as input argument but I want to pass the variable having the original image as input .

Are there any function that can allow me to do that. As it has to be the variable that I will be passing as I wont be getting the image file name.

I have already tried

x, img = data.transforms.presets.ssd.load_test(im_fname, short=512)

where im_fname is the file name of the image in (300,300,3) format . but I cant pass the file name as the images will be going from a ImageDataGenerator into a preprocessing func where this yolo net will be used.

Answer 1

With numpy , you can simply change the image from channel last to channel first using the following code:

import numpy as np

img = np.random.randn(300, 300, 3)  # creating random image

print(img.shape) #Out: (300, 300, 3)

img_channel_first = np.moveaxis(img, -1, 0)

print(img_channel_first.shape) #Out: (3, 300, 300)

img_in = img_channel_first[np.newaxis,:]

print(img_in.shape) #Out: (1, 3, 300, 300)