在推理 tensorflow 2 之前获取元素检测 [英] Get element detections before the inference tensorflow 2

问题描述

这周我在玩"使用 Tensorflow 2，我尝试对象检测，但我不知道如何执行以下操作:

this week i'm "playing" with Tensorflow 2 and i try object detection and i dont know how to do the following:

在教程 TF2 object detection 中，获取一张图像中某些元素的推断，如以下代码所示:

In the tutorial TF2 object detection, get the inference of some elements in one image, as i show in the following code:

  image_np = load_image_into_numpy_array(image_path)
  input_tensor = tf.convert_to_tensor(image_np)
  input_tensor = input_tensor[tf.newaxis, ...]
  detections = detect_fn(input_tensor)

但我需要在推理之前检测到元素或区域.我的意思是，建议区域的坐标，但我不知道该怎么做.我尝试拆分过程，一方面是区域提议，另一方面是推理.

But i need to get the elements or regions detected, before the inference. I mean, the coordinates of the proposed regions but i dont know how to do that. I try to split the process, in one hand the region proposal and in the other hand the inference.

我的代码如下:

def make_inference(image_path,counter,image_save):
  print('Running inference for {}... '.format(image_path), end='')
  image_np = load_image_into_numpy_array(image_path)
  input_tensor = tf.convert_to_tensor(image_np)
  input_tensor = input_tensor[tf.newaxis, ...]
  detections = detect_fn(input_tensor)
  num_detections = int(detections.pop('num_detections'))
  detections = {key: value[0, :num_detections].numpy()
                   for key, value in detections.items()} 
  detections['num_detections'] = num_detections
  detections['detection_classes'] = detections['detection_classes'].astype(np.int64)
  image_np_with_detections = image_np.copy()
  viz_utils.visualize_boxes_and_labels_on_image_array(
        image_np_with_detections,
        detections['detection_boxes'],
        detections['detection_classes'],
        detections['detection_scores'],
        category_index,
        use_normalized_coordinates=True,
        max_boxes_to_draw=200,
        min_score_thresh=.5,
        agnostic_mode=False)
  plt.axis('off')
  plt.imshow(image_np_with_detections)
  nombre = str(counter)+'.jpg'
  plt.savefig('/content/RESULTADOS/'+nombre,  dpi=dpi ,bbox_inches='tight')
  counter = counter+1
  plt.clf()

提前致谢.

推荐答案

我是一名软件工程师，数据科学确实需要实现很多 OOPS(但是 Python 中的 OOPS 是一个笑话 [IMO])，我已经采取了可以自由地绘制一个类，并具有以下功能来获取 List[DetectedObj]

I worked as a software engineer and data science really requires a lot of OOPS implemented hence (however OOPS in Python is a Joke [IMO]), I have taken the liberty to draw out a class instead and have the following function to get a List[DetectedObj]

简单的 POJO 类，用于保存您收到的每个检测.

Simple POJO class to hold every detection you received.

from typing import Dict, Any, Optional, List
import numpy as np

class DetectedObject:
 
        def __init__(self, ymin: float, xmin: float, ymax: float, xmax: float, category: str, score: float):
               self.xmin = xmin
               self.ymin = ymin
               self.xmax = xmax
               self.ymax = ymax
               self.clazz = clazz
               self.score=score

调用以下函数&通过您从detect_fn收到的检测

Call the following function & pass your detections which you received from detect_fn

def get_objects_from_detections(detections: Dict[str, Optional[Any]], categories: Dict[int, Optional[Any]], threshold: float = 0.0) -> List[DetectedObject]: 
        det_objs = []
        bbox_list = detections['detection_boxes'].tolist()
        for i, clazz in np.ndenumerate(detections['detection_classes']):
            score = detections['detection_scores'][i]
            if score > threshold:
                clazz_cat = categories[clazz]['name']
                row = bbox_list[i[0]]
                tiny = DetectedObject(row[0], row[1], row[2], row[3], clazz_cat, score)
                det_objs .append(tiny)
        return det_objs

这篇关于在推理 tensorflow 2 之前获取元素检测的文章就介绍到这了，希望我们推荐的答案对大家有所帮助，也希望大家多多支持IT屋！