Performance issue in the definition of libs/networks/efficientnet/condconget_condconv_initializer, v/condconv_layers.py(P1)

Thinklab-SJTU / R3Det_Tensorflow

Code for AAAI 2021 paper: R3Det: Refined Single-Stage Detector with Feature Refinement for Rotating Object

Apache License 2.0

540 stars 122 forks source link

Performance issue in the definition of libs/networks/efficientnet/condconget_condconv_initializer, v/condconv_layers.py(P1) #136

Open DLPerf opened 3 years ago

DLPerf commented 3 years ago

Hello, I found a performance issue in the definition of get_condconv_initializer, libs/networks/efficientnet/condconv/condconv_layers.py, tf.reshape(kernel, [-1]) will be calculated repeatedly during program execution, resulting in reduced efficiency. I think both it and kernel should be created before the loop in condconv_initializer.

Looking forward to your reply. Btw, I am very glad to create a PR to fix it if you are too busy.

DLPerf commented 3 years ago

@yangxue0827 Hi, my friend, could you consider my issue asap?