- flash PYNQ V2.5 to the SD card
- upgrade the board using https://github.com/Xilinx/DPU-PYNQ
- Rename the util-modelXX.py to utils.py and put it on the board together with model file: modelXX.elf.
- Run test_accuracy_latency.ipynb to get the latency and accuracy of the tested model.
- Change the threadnum in utils.py to use different number of threads.
The ImageNet validation set should be put under ./val which contains ILSVRC2012_val_00000001.JPEG, ILSVRC2012_val_00000002.JPEG, ..., ILSVRC2012_val_00050000.JPEG.
All the model weights and activations are quantized to 8 bits.