0

I'm training ssd_mobilenet_v1_coco on a custom dataset using the Tensorflow Object Detection API with 15000 steps on Google Colab. Now every time I train, only 3 checkpoints get saved. I want to increase the number of saved checkpoints. I tried modifying \models\research\object_detection\legacy\trainer.py file like below

saver = tf.train.Saver(keep_checkpoint_every_n_hours=keep_checkpoint_every_n_hours,max_to_keep = 30)

But then it stores 5 checkpoints.

Is there any way I can save like 10/15 checkpoints or more?

zahin178
  • 13
  • 1
  • 1
  • 3
  • Did you tried [this](https://stackoverflow.com/a/66027554/14290681)? This may help you. –  Sep 07 '21 at 11:31
  • Thanks for the link @TFer2. But the problem is my training takes about 40 minutes and the checkpoints are saved around 10 minutes. So, even if I try to save the previous checkpoints I will only save 3/5. Can you suggest me how I can change the frequency of saving? I mean how to save a checkpoint after every 3 minutes instead of 10? – zahin178 Sep 08 '21 at 13:57

0 Answers0