tiny-yolo on MAIX


#1

According to https://www.indiegogo.com/projects/sipeed-maix-the-world-first-risc-v-64-ai-module/x/20442306#/, the MAIX supports tiny-yolo. However, I couldn’t find any further information in the documentation. Are there any examples on how to get it runnning? Thanks!


#2

Hi Tobias - I’ve not gotten that far yet with my board, but I believe what you are looking for can be found here: https://github.com/kendryte/kendryte-standalone-demo/tree/develop/kpu.

There are some good notes also about how to build and load models (using yolo as an example) in the model-compiler repository as well: https://github.com/kendryte/kendryte-model-compiler/blob/master/How%20to%20Use%20Kendryte%20Model%20Compiler.md

Note that these are generic Kendryte chip repositories, so some tweaking to match your specific MAIX peripheral config might be required.


#3

Hi again - After doing a bit more exploring, I believe the “ai_demo_sim” example is probably using a yolo cnn model as well: https://github.com/sipeed/LicheeDan_K210_examples/tree/master/src/ai_demo_sim

If you just want to see something working and don’t care about tweaking or retraining then probably building and installing this would be your quickest option.


#4

Hi,please refer to https://maixpy.sipeed.com/zh/libs/Maix/kpu.html
you can use micropython to run yolo2!


#5

The example kpu-examples work quite well, but the current maixpy build (http://dl.sipeed.com/MAIX/MaixPy/release/maixpy_v0.1.1_beta.bin) doesn’t include the kpu module. In order to run the examples on your board, you need to build the bin yourself. Follow the thread: [ HELP ] unrecognized opcode `fmv.d.x f0,x0' in order to use the correct toolchain, otherwise you won’t be able to build the bin.


#6

hi, here is the new firmware support kpu: MaixPy Firmware update (MaixPy 固件更新贴)


#7

Ah, I didn’t see this - it would be good to also link this binary in https://github.com/sipeed/MaixPy/releases for all users which follow https://maixpy.sipeed.com/zh/get_started/upgrade_firmware.html


#8

Thanks, @mwilkinson, @Zepan, and @frdnd for the help. I upgraded to firmware maixpy_beta_0.2.0.zip and tried to run the examples given on https://maixpy.sipeed.com/zh/libs/Maix/kpu.html. However, the model address (offset) seems to be wrong:

>>> task = kpu.load(0xd00000)
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
ValueError: [MAIXPY]kpu: load error -2

Any ideas? Thanks!


#9

Hi, please upgrade to newest 0.2.4 firmware in github