search

doonny / PipeCNN

An OpenCL-based FPGA Accelerator for Convolutional Neural Networks
Apache License 2.0
1.26k stars 369 forks source link

how to config addition_precision_config in layer_config.h #158

Open WenyeLiu opened 3 years ago

WenyeLiu commented 3 years ago

Hi,

I am going to deploy a resnet like model on pipecnn. However, the accuracy is extremently low comparing with the results in the simulation framework. I underestand that the precision_config is used to set the quantiazation bitwidth for weights and input/output feature maps. May I ask what is the usage of the "addition_precision_config" in layer_config.h. It seems used to convert between char and float data type for batchnorm layer and eltwise layer. Is there any explanation for those parameters in "addition_precision_config"

Many thanks.

doonny commented 3 years ago

will update the instructions later

WenyeLiu commented 3 years ago

Thanks doonny. Looking forward your tips.