How to convert TF object detection models to tf.keras?

Asked May 09 '21 at 20:34

Active May 09 '21 at 20:34

Viewed 126 times

Is there a way to convert TF object detection models from here to tf.keras? If not, how can I get the total number of variables in the model?

asked May 09 '21 at 20:34

balu

have you looked into their collab ? https://github.com/tensorflow/models/blob/master/research/object_detection/colab_tutorials/eager_few_shot_od_training_tf2_colab.ipynb – David Thery May 09 '21 at 21:15
Thank you for your suggestion, at least I was able to count the number of weights. But I'm not sure I'm doing well. I'm running a MobileNet V2 SSD FPNLite 320x320 model and I get around 2.6 million of parameters while in the MobileNet paper https://arxiv.org/pdf/1801.04381.pdf they report a value of 4.3 million for SSDLite (and I also have the FPN subnet). – balu May 10 '21 at 07:11
Hi @balu ! Could you check this thread for getting number of parameters from trainable_variables. https://stackoverflow.com/questions/38160940/how-to-count-total-number-of-trainable-parameters-in-a-tensorflow-model – Nov 05 '21 at 02:10

0 Answers0