what does num_epochs means in eval_input_config TF2 object detection API v2

user_8912375 0
Viewed 440

what does num_epochs means in the eval_input_config in the pipeline.config I noticed that when I change the value of this field to more than 1 I'm having these warning logs when trying to train an object detection model using Tensorflow Object Detection API 2

[04/14 22:39:04] tensorflow WARNING: Ignoring detection with image id 163697061 since it was previously added 

eval_input_reader: {
  label_map_path: "PATH_TO_BE_CONFIGURED/label_map.txt"
  shuffle: false
  num_epochs: 1
  tf_record_input_reader {
    input_path: "PATH_TO_BE_CONFIGURED/val2017-?????-of-00032.tfrecord"
  }
tensorflow-model-garden object-detection-api object-detection tensorflow2.0 python
1 Answers

Normally, one epoch is a single pass through your entire training set,and you would need to run multiple epochs until the model converges while training.
However, since in this case, it's inside the eval_input_reader, it should be 1 because for validation, you need to go through the test/validation set only once. It doesn't make sense to go over the validation set multiple times as the weights don't get updated/no backpropagation happens

user_100376890
Related