When trying to compile amodel for the DPU you usually have to freeze the model, quantisize and then compile.
Often freezing and quantisizing a model works fine, but the compilation throws really often errors.
My questions are:
- What do you prefer to use when trying to infer a model on the FPGA TensorFlow or Caffe?
- What are the pros and cons?
What I am actually try to do is to bring PointNet on the FPGA. But I ran in several problems, when trying to compile the Model for the DPU. It is quite frustrating.
If there are any previous attempts to compile point cloud classfication network to the DPU. Please tell me. I didn’t find any.