MLIR Inference Code

import cv2
import onnx
import numpy as np
import os
import onnxruntime
import time


def get_image(image_dirs) :
    image_list = []
    for roots, dirs, files in os.walk(image_dirs) :
        for file in files :
            image_dir = os.path.join(roots, file)
            if image_dir.endswith("png") or image_dir.endswith("jpg") or image_dir.endswith("JPEG") :
                image_list.append(image_dir)

    return image_list



def preprocess_image(image_path) :
    image_path = image_path
    image = cv2.imread(image_path, 1) #image read
    image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)
    image_data = cv2.resize(image, (256, 256), interpolation=cv2.INTER_LINEAR).astype(np.float32) # cv2.INTER_LINEAR
    image_data = image_data[16:240, 16:240, :].copy()
    
    image_data = image_data.transpose([2, 0, 1]) # C, H, W
    mean = np.array([0.079, 0.05, 0]) + 0.406 # 0.485, 0.456, 0.406
    std = np.array([0.005, 0, 0.001]) + 0.224 #0.229, 0.224, 0.225

    for channel in range(image_data.shape[0]): 
        image_data[channel, :, :] = (image_data[channel, :, :] / 255 - mean[channel]) / std[channel] #RGB

    image_data = np.expand_dims(image_data, axis=0)
    
    return image_data


def softmax(x):
    e_x = np.exp(x - np.max(x))
    return e_x / e_x.sum()


with open("imagenet_classes.txt", "r") as f:
    categories = [s.strip() for s in f.readlines()]


model_path = "./mobilenetv2-7.onnx"


data = "./parrot.jpg"
#image = get_image(data)
input_data = preprocess_image(data)


session = onnxruntime.InferenceSession(model_path)
input_name = session.get_inputs()[0].name
output_name = session.get_outputs()[0].name


output = session.run([output_name], {input_name:input_data})[0]
output = output.flatten()
output = softmax(output) # this is optional

top1 = np.argmax(output)

print(f"{categories[top1]},{output[top1]}")

import numpy as np
from PyRuntime import OMExecutionSession
import cv2

with open("imagenet_classes.txt", "r") as f :
    categories = [s.strip() for s in f.readlines()]


def softmax(x):
    e_x = np.exp(x - np.max(x))
    return e_x / e_x.sum()

def get_image(image_dirs) :
    image_list = []
    for roots, dirs, files in os.walk(image_dirs) :
        for file in files :
            image_dir = os.path.join(roots, file)
            if image_dir.endswith("png") or image_dir.endswith("jpg") or image_dir.endswith("JPEG") :
                image_list.append(image_dir)

    return image_list

def preprocess_image(image_path) :
    image_path = image_path
    image = cv2.imread(image_path, 1) #image read
    image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)
    image_data = cv2.resize(image, (256, 256), interpolation=cv2.INTER_LINEAR).astype(np.float32) # cv2.INTER_LINEAR
    image_data = image_data[16:240, 16:240, :].copy()
    
    image_data = image_data.transpose([2, 0, 1]) # C, H, W
    mean = np.array([0.079, 0.05, 0]) + 0.406 # 0.485, 0.456, 0.406
    std = np.array([0.005, 0, 0.001]) + 0.224 #0.229, 0.224, 0.225

    for channel in range(image_data.shape[0]): 
        image_data[channel, :, :] = (image_data[channel, :, :] / 255 - mean[channel]) / std[channel] #RGB
        
    image_data = np.expand_dims(image_data, axis=0)
    image_data = np.ascontiguousarray(image_data.data)

    return image_data


model = './mobilenetv2-7.so' # LeNet from ONNX Zoo compiled with onnx-mlir

data = "./parrot.jpg"
#image = get_image(data)
input_data = preprocess_image(data)


session = OMExecutionSession(shared_lib_path=model)

print("input signature in json", session.input_signature())
print("output signature in json",session.output_signature())
print(input_data.shape)
outputs = session.run(input=[input_data])


outputs = softmax(outputs[0][0])
idx = np.argmax(outputs)
print(f"Label: {categories[idx]}, score: {outputs[idx]}")

import torch
from PIL import Image
from torchvision import transforms
import argparse

def main():
    parser = argparse.ArgumentParser(formatter_class=argparse.ArgumentDefaultsHelpFormatter)

    parser.add_argument(
      '-m', '--image', required=True, help='File path of image file.')
    parser.add_argument(
      '-l', '--label', help='File path of label file.')

    args = parser.parse_args()

    filename = args.image
    label = args.label

    # 카테고리 읽어들이기
    # with open("imagenet_classes.txt", "r") as f:
    #     categories = [s.strip() for s in f.readlines()]
    #filename = "./parrot.jpg"

    #카테고리 읽어들이기
    with open(label, "r") as f:
        categories = [s.strip() for s in f.readlines()]


    preprocess = transforms.Compose([
        transforms.Resize(256),
        transforms.CenterCrop(224),
        transforms.ToTensor(),
        transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]),
    ])

    input_image = Image.open(filename)

    input_tensor = preprocess(input_image)
    input_batch = input_tensor.unsqueeze(0) # 모델에서 요구하는 형태의 미니배치 생성

    model = torch.hub.load('pytorch/vision:v0.10.0', 'mobilenet_v2', pretrained=True)
    model.eval()

    if torch.cuda.is_available():
        input_batch = input_batch.to('cuda')
        model.to('cuda')

    with torch.no_grad():
        output = model(input_batch)

    print(output[0])
    probabilities = torch.nn.functional.softmax(output[0], dim=0)
    #print(probabilities)

    top5_prob, top5_catid = torch.topk(probabilities, 5)
    for i in range(top5_prob.size(0)):
        print(f"{categories[top5_catid[i]]} {top5_prob[i].item()}")


if name == 'main':
    main()