model_processor.py

import os
import cv2
import numpy as np
import argparse
import sys

sys.path.append('../')

from acl_model import Model
from image_net_classes import get_image_net_class



class ModelProcessor:
    
    def __init__(self, acl_resource, params):
        self._acl_resource = acl_resource
        self.params = params
        self._model_width = params['width']
        self._model_height = params['height']

        assert 'model_dir' in params and params['model_dir'] is not None, 'Review your param: model_dir'
        assert os.path.exists(params['model_dir']), "Model directory doesn't exist {}".format(params['model_dir'])
            
        # load model from path, and get model ready for inference
        self.model = Model(acl_resource, params['model_dir'])

    def predict(self, img_original):
        
        #preprocess image to get 'model_input'
        model_input = self.preprocess(img_original)

        # execute model inference
        infer_output = self.model.execute([model_input]) 

        # postprocessing: 
        category = self.post_process(infer_output)

        return category

    def preprocess(self,img_original):
        '''
        preprocessing: resize image to model required size, and normalize value
        '''
        scaled_img_data = cv2.resize(img_original, (self._model_width, self._model_height))
        normalized_img = scaled_img_data - np.array([123,117,104])
        # Caffe model after conversion, need input to be NCHW; the orignal image is NHWC, need to be transposed (use .copy() to change memory format)
        preprocessed_img = np.asarray(normalized_img, dtype=np.float16).transpose([2,0,1]).copy()
        
        return preprocessed_img

    def post_process(self, infer_output):
        print("post process")
        data = infer_output[0]
        vals = data.flatten()
        top_k = vals.argsort()[-1:-6:-1]
        print("======== top5 inference results: =============")
        for n in top_k:
            object_class = get_image_net_class(n)
            print("label:%d  confidence: %f, class: %s" % (n, vals[n], object_class))
        
        object_class = get_image_net_class(top_k[0])
        return object_class