Get element detections before the inference tensorflow 2 - tensorflow

Question

this week i'm "playing" with Tensorflow 2 and i try object detection and i dont know how to do the following:
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)
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.
My code is the following:
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()
Thanks in advance.

Answer 1

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]
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
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