Serveless

Author

Alfa Pradana

Dataset

This work is based on the previous one. Need to download the data here:

Library & function

import numpy as np
import onnxruntime as ort

from io import BytesIO
from urllib import request

from PIL import Image

def download_image(url):
    with request.urlopen(url) as resp:
        buffer = resp.read()
    stream = BytesIO(buffer)
    img = Image.open(stream)
    return img

def prepare_image(img, target_size):
    if img.mode != 'RGB':
        img = img.convert('RGB')
    img = img.resize(target_size, Image.NEAREST)
    return img

Load model

onnx_model_path = "../data/hair_classifier_v1.onnx"
session = ort.InferenceSession(onnx_model_path, providers=["CPUExecutionProvider"])

inputs = session.get_inputs()
input_name = inputs[0].name
input_name

'input'

outputs = session.get_outputs()
output_name = outputs[0].name
outputs[0]

<onnxruntime.capi.onnxruntime_pybind11_state.NodeArg at 0x2c680022270>

Image size

img_url = "https://habrastorage.org/webt/yf/_d/ok/yf_dokzqy3vcritme8ggnzqlvwa.jpeg"
img = download_image(img_url)

x = np.array(img)
x.shape

(1024, 1024, 3)

Resize

img_rz = prepare_image(img, (200, 200))
img_rz

Preprocessing

x = np.array(img_rz)

x = x / 255.0 # scale to 0-1 (like ToTensor does)

# normalize with ImageNet mean and std
mean = np.array([0.485, 0.456, 0.406])
std = np.array([0.229, 0.224, 0.225])
x = (x - mean) / std

x[0, 0, 0]

-1.0732939463995204

Model output

# from (Height, Width, Channel) to (C, H, W) for PyTorch format
# (200, 200, 3) -> (3, 200, 200)
x = np.transpose(x, (2, 0, 1))

# For batch dimension: (C, H, W) -> (1, C, H, W)
# (3, 200, 200) -> (1, 3, 200, 200)
x = np.expand_dims(x, axis=0)

# Convert to float32
x = x.astype(np.float32)

result = session.run([output_name], {input_name: x})
result[0]

array([[0.08927415]], dtype=float32)

Work with docker

docker images

Model output docker

docker run -it -rm --entrypoint=base hw09 -c "python q6.py"