1

I want to quantize the ssd-mobilenet model , then implementation on FPGA， now i use the ssd_mobilenet_v1_quantized_coco model http://download.tensorflow.org/models/object_detection/ssd_mobilenet_v1_quantized_300x300_coco14_sync_2018_07_18.tar.gz ,then convert .pb file to .tflite file ,then can i use the tflite model on FPGA?

lx d
  • 31
  • 3
  • After training tflite model with post training quantization and viewing the model using Netron, i figured out that some first layers are still in float! Which indeed needs multiple FPGA kernels to be implemented which is not efficient. – robert Sep 01 '19 at 11:34

1 Answers1

0

Currently Tensorflow only Supports android, ios and microcontrollers. as you need to synthese Hardware blocks in FPGA, you need to write inference of you network manually in c++ or rtl.

robert
  • 1
  • 2
  • That's what I am trying to do, but can not get the calculations implemented correctly. Could you please take a look at my question [here](https://stackoverflow.com/questions/59118407/calculation-operations-with-the-parameters-of-a-tflite-quantized-model) – Nazar Nov 30 '19 at 21:11