Postprocessing method for exported model?

[3zhang]

I'm following this tutorial for object detection:
https://www.tensorflow.org/tfmodels/vision/object_detection

At the end, the Visualize predictions part, the result visualized are based on the resized input. Is there any postprocessing function to convert the resized result into the result corresponding to the original image?

#Visualize predictions.
input_image_size = (HEIGHT, WIDTH)
plt.figure(figsize=(20, 20))
min_score_thresh = 0.30 # Change minimum score for threshold to see all bounding boxes confidences.

for i, serialized_example in enumerate(test_ds):
  plt.subplot(1, 3, i+1)
  decoded_tensors = tf_ex_decoder.decode(serialized_example)
  image = build_inputs_for_object_detection(decoded_tensors['image'], input_image_size)
  image = tf.expand_dims(image, axis=0)
  image = tf.cast(image, dtype = tf.uint8)
  image_np = image[0].numpy()
  result = model_fn(image)
  visualization_utils.visualize_boxes_and_labels_on_image_array(
      image_np,
      result['detection_boxes'][0].numpy(),
      result['detection_classes'][0].numpy().astype(int),
      result['detection_scores'][0].numpy(),
      category_index=category_index,
      use_normalized_coordinates=False,
      max_boxes_to_draw=200,
      min_score_thresh=min_score_thresh,
      agnostic_mode=False,
      instance_masks=None,
      line_thickness=4)
  plt.imshow(image_np)
  plt.axis('off')

plt.show()

[laxmareddyp]

Hi @3zhang ,

There is no mention of a specific postprocessing function to convert the resized result back to the original image in the TensorFlow object detection tutorial. However, it's likely because the bounding boxes coordinates predicted by the model are based on the resized image. You can simply scale the bounding boxes back to the original image dimensions to get the corresponding bounding boxes in the original image.

As a work around,Calculate the scaling factor between the original image size and the resized image size.
Multiply the predicted bounding box coordinates (xmin, ymin, xmax, ymax) by the scaling factor to get the coordinates in the original image.

Meanwhile we will try modify the above visualization function with few parameters changes and check the results on original image input directly instead of resized image.

Thanks.

[github-actions]

This issue has been marked stale because it has no recent activity since 7 days. It will be closed if no further activity occurs. Thank you.

[github-actions]

This issue was closed due to lack of activity after being marked stale for past 7 days.

[google-ml-butler]

Are you satisfied with the resolution of your issue?
Yes
No