error while running eval.py tensorflow object detection api

Question

When I run the eval.py script, the images are evaluated and in the logs I can see that visualizations are also created on the images.

I have set num_examples to 50 in the pipeline.config. However, 50 images are not evaluated. After evaluation of some 9 images (which varies a lot - sometimes 5, sometimes 4), I get an error that ValueError: Image with id 1531471339_visible.png already added.

I am not sure where I am going wrong.

Note: This is my own dataset that I've trained. I also tried different machines, and still the same error.

eval config:

eval_config {
  num_examples: 50
  use_moving_averages: false
}
eval_input_reader {
  label_map_path: "/home/path/to/labelmap.pbtxt"
  tf_record_input_reader {
    input_path: "/home/path/to/file.tfrecord"
  }
}

Answer 1

I got the same error message. I solved it by adding

metrics_set: "coco_detection_metrics"

into my config file.

eval_config {
  num_examples: 50
  use_moving_averages: false
  metrics_set: "coco_detection_metrics"
}

Then the error message disappear... not sure why. Maybe you can give it a go. Let me know if it helps.