如何使用 TensorFlow 裁剪圖像？

Question

我正在嘗試裁剪檢測到 object 的圖像。 從 TensorFlow 的文檔中有一個 function。

tf.image.crop_to_bounding_box(image, offset_height, offset_width, target_height, target_width)

我正在嘗試弄清楚如何獲取給定的 arguments 但不確定要使用哪些信息。 這是我正在使用的代碼。

img = cv2.imread(IMAGE_PATH)
image_np = np.array(img)

input_tensor = tf.convert_to_tensor(np.expand_dims(image_np, 0), dtype=tf.float32)
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
# detection_classes should be ints.
detections['detection_classes'] = detections['detection_classes'].astype(np.int64)

label_id_offset = 1
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']+label_id_offset,
            detections['detection_scores'],
            category_index,
            use_normalized_coordinates=True,
            max_boxes_to_draw=9,
            min_score_thresh=.5,
            agnostic_mode=False)

#print(detections['detection_boxes'])
plt.imshow(cv2.cvtColor(image_np_with_detections, cv2.COLOR_BGR2RGB))
plt.show()

結果就是這樣，我試圖裁剪除邊界框之外的所有內容 -已識別 object

Answer 1

如果問題是為了獲得解決方案來裁剪邊界框的內容並丟棄它之外的所有內容，則應該像這樣工作：

# ’image’ is the input image tensor
boxes = detections['detection_boxes']
cropped_boxes = []

# this currently crops all bounding boxes, not taking into account the scores
for i in range(boxes.shape[0]):
    box = tuple(boxes[i].tolist())
    ymin, xmin, ymax, xmax = box 
    cropped_boxes.append(
        tf.image.crop_to_bounding_box(
            image=image, 
            offset_height=ymin, 
            offset_width=xmin, 
            target_height=ymax-ymin, 
            target_width=xmax-xmin
        ))

有關如何解釋框的信息可以從用於繪制結果的問題中使用的方法visualize_boxes_and_labels_on_image_array的源代碼中找到。

https://github.com/tensorflow/models/blob/457bcb8595903331932e2faf95bec8ba69e04688/research/object_detection/utils/visualization_utils.py