So far I have taken one mnist image, and done the following:

` from PIL import Image
import numpy as np

def preprocess(img, format, dtype, h, w, scaling):

sample_img = img.convert('L')


resized_img = sample_img.resize((1, w*h), Image.BILINEAR)
resized = np.array(resized_img)
resized = resized.astype(dtype)

return resized

png img file

img = Image.open('./7.png')

preprocessed img, FP32 formated numpy array

img = preprocess(img, format, "float32", 28, 28, None)

to bytes

img = img.tofile('7.binaryimage',format='binary') curl -X POST 192.168.34.174:8000/v2/models/keras_mnist/versions/1 -H "Content-Type: application/octet-stream" -H' NV-InferRequest:batch_size: 1 input { name: "dense_input" dims: [-1, 784] } output { name: "activation_2" cls { count: 1 } }' --data-binary "@7.binaryimage" `