Pytorch --> ONNX --> Tensorflow. How to translate the Bounding Box Voting part?
MroutsideturnAF opened this issue · 3 comments
MroutsideturnAF commented
When transforming the model to Onnx and then to Tensorflow. It returns output in the shape of 3 Tensors:
loaded model outputs = [<tf.Tensor 'Identity:0' shape=(1, 16800, 4) dtype=float32>, <tf.Tensor 'Identity_1:0' shape=(1, 16800, 2) dtype=float32>, <tf.Tensor 'Identity_2:0' shape=(1, 16800, 10) dtype=float32>]
This format corresponds to loc, conf, landms = net(img) # forward pass
from detect.py
Is there a recommendable way to translate the bounding box part to Tensorflow aswell?
alicera commented
Hi AndyMcPython ,
May I ask your network learning for resnet50 and mobilenet?
MroutsideturnAF commented
Hi @alicera,
I use the resnet50.
alicera commented
How to adjust the learning rate for better training?