This repository contains the code used to deploy NanoFlowNet on the AI-deck.
Please refer to the main repository for the used models and the Crazyflie obstacle avoidance application
git clone [email protected]:GreenWaves-Technologies/gap_sdk.git
git checkout ef8ee923dfb39cf2ef962d25e5bf69f8881ebfb3
Implement the following bug workaround in gap_sdk
. Modify rtos/pmsis/bsp/include/bsp/ai_deck.h
, put the following at the bottom of the file, before the #endif
// #define CONFIG_SPIFLASH_SECTOR_SIZE (1<<12)
#include "bsp/flash/hyperflash.h"
#include "bsp/ram/hyperram.h"
#define pi_default_flash_conf pi_hyperflash_conf
#define pi_default_flash_conf_init pi_hyperflash_conf_init
#define pi_default_ram_conf pi_hyperram_conf
#define pi_default_ram_conf_init pi_hyperram_conf_init
// #endif
Follow the installation instructions (README.md, make sure you are on the suggested commit). Source the AI-deck config file; configs/ai_deck.sh
. Modify OPENOCD_CABLE
if necessary (redo source). Perform a full install; make sdk
.
nntool_script
can be modified to include grayscale image (112hx160w) file paths for quantization and quantization evaluation. No sample images are included. Default config uses fquant ("fake" quantization).
To execute on the AI-deck, in this repo:
make all run PMSIS_OS=pulpos platform=board
To flash
make all flash PMSIS_OS=pulpos platform=board
freertos
and gvsoc
are not supported.
Please cite us as follows:
@inproceedings{bouwmeester2023nanoflownet,
title={Nanoflownet: Real-time dense optical flow on a nano quadcopter},
author={Bouwmeester, Rik J and Paredes-Vall{\'e}s, Federico and De Croon, Guido CHE},
booktitle={2023 IEEE International Conference on Robotics and Automation (ICRA)},
pages={1996--2003},
year={2023},
organization={IEEE}
}