Performance issues in training_api/research/object_detection/utils/ops.py
JamesCao2048 opened this issue · 2 comments
Hello, I found that in training_api/research/object_detection/utils/ops.py, tf.zeros andtf.shapeare repeatedly created in the loop, here. As they ratain same in each iteration, I think they should be moved before the loop to avoid creating redundant nodes in the tf computation graph.
Looking forward to your reply. Btw, I am very glad to create a PR to fix it if you are too busy.
@Daniel-Jess Hello, bro, I hope you could deal with the issue asap.
@JamesCao2048 Hello, regarding you issue, the file you are mentioning is based on TF1 Object detection API without any modification link
It would be helpful if you can raise this issue to the TensorFlow git repository and then if this modification is accepted, then we can make adjustments to the current code.