ManishSoni1908 / Mobilenet-ssd-keras

A Keras port of Single Shot MultiBox Detector as Mobilenet as a Backbone
62 stars 24 forks source link

Getting Nan Values while training from scratch #21

Open ManishKakarla opened 1 year ago

ManishKakarla commented 1 year ago

Working fine for 3 batches in first iteration after that the loss was becoming Nan or inf.Can anyone had faced a similar issue or know where the issue is please help me,.

WARNING:tensorflow:`epsilon` argument is deprecated and will be removed, use `min_delta` instead.
Epoch 1/10
  3/234 [..............................] - ETA: 33:37:24 - loss: 44.4110Batch 3: Invalid loss, terminating training
  4/234 [..............................] - ETA: 25:09:00 - loss: inf