1

I would like to be able to visualize the regions proposals made by Faster-RCNN (like Resnet101_coco) using Tensorflow Object Detection API, preferably in Tensorboard. Is there any way to do so ?

jood
  • 2,188
  • 2
  • 21
  • 32

1 Answers1

0

You can visualize the detected objects in tensorboard during evaluation (i.e. while running object_detection/eval.py script) You would need to add num_visualizations key to the config file, e.g.

eval_config: {
  num_examples: 20000
  num_visualizations: 16
  min_score_threshold: 0.15
  # Note: The below line limits the evaluation process to 10 evaluations.
  # Remove the below line to evaluate indefinitely.
  max_evals: 1
}

After running evaluation, you should be able to see an images tab in Tensorboard showing the detected objects. You can adjust the IoU threshold (min_score_threshold) to vary the number of displayed detections.

Vadim Smolyakov
  • 1,187
  • 11
  • 24
  • Yes, but you are talking about how to see more examples in tensorboard. I'm not trying to see more final outputs, i'm trying to see the region proposals (before the classification network) – jood Aug 09 '18 at 08:23
  • Ah! I see, the internal region proposals? My answer relates to the final detections. – Vadim Smolyakov Aug 09 '18 at 15:50
  • Yes! I actually asked this question here: https://stackoverflow.com/questions/50027635/show-more-images-in-tensorboard-tensorflow-object-detection :) – jood Aug 16 '18 at 09:03