New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We鈥檒l occasionally send you account related emails.
Already on GitHub? Sign in to your account
Upgrade to SDK Version 3.7 #23
Conversation
Hi! Thanks for all of this! I'm currently myself on a holiday so it might take some time to review this pull request but I can check if I have some time these days to try it out. Since the SDK is evolving so quickly, it might be good to maybe just stick with one SDK for a little while maybe? That is kind of the reason why I made the docker files in the first place, since it takes a lot of time to update all of this and we are busy with many other projects as well. But I'm happy that you are contributing as much as you do, it is definitely helping out a lot! The blocking problem I noticed in SDK 3.6 as well. So this only affects the streaming I guess? I heard from @Yaooooo that there is now an option to do the streaming compression from the cluster instead, so maybe that is something worth looking into? I see that you are still pushing many changes to your PR so let me know if you are finished. |
Hey @knmcguire, Regarding sticking to one version I fully agree with you. What do you think, should we stick with Regarding the streaming, that sounds great. But in other words it's either to use an older version (like |
It sounds good to stick with 3.7 for a while. For the streaming we should probably advise people to use the 3.5 SDK (since that is the one that still works). I've already added instructions in the readme how to make an image for previous versions as well, since that is the whole reason why I started to use docker so that I can easily switch between gapsdk:3.4, 3.5 etc :) By the way, the neural network examples are indeed very old. If mnist_simple is completely broken than we should just remove it actually. It is not really platform depended anyway. But surprised that the gap_flow_extern seems to work in 3.7. I can't remember if it actually uses a working neural network but was more of an example on how to compile an empty network on the chip. I'll look into it again once I get back :) |
Alright! I just checked the streaming examples with Yes removing the Afaik |
Hey @knmcguire, |
Hi! I'll be back next week so I've already started looking into it, so I'll have some comments for you soon. It would be good to start merging this into the code indeed to not keep it lingering around in PR limbo :) It has become a pretty big pull request btw, but I won't be very picky since this is a bit of an experimental repo anyway for now! |
I tried out with the new docker file + instructions and it worked right away on the image manipulations example :) Just a couple of things of things that I have tested
I haven't gone through the gapflow example yet or worked on the streaming examples. |
The problem is/was that the demosaicking function is placed in the common folder at the root of the GAP8-examples which is not mounted in the docker. But since the file is not used anywhere else I moved it in the test camera example.
Good point, I added a small paragraph about it |
So I tested out the send uart, that still works! The gapflow_model_extern I get the following error when I use docker.
But when I run docker with I got from @Yaooooo some instructions as well to make it completely standalone, but maybe best for a next pull request ;) |
Good to hear that the UART example still works.
I put this at the end of the readme inside the |
Yeah maybe for now it is enough to include the .mk files from common and add it to the the gapflow_model_extern |
Done that now :) |
ah perfect! I was actually also right in the middle of doing this but for some reason I removed my SDK 3.7 image from docker so I had to build it again :P Probably I'm going to start moving the readmes of docker and the gapflow readme to the doc folder. I think it is stable enough to also let other people know that they can use it. |
Hey,
I took some time and updated most of the examples to the latest SDK Version (
3.7
) for the GAP8. Not everything is perfected yet but since the SDK is also evolving quite quick it's hard to keep everything up-to-date.test_functionalities/test_camera
still remainsis better but not optimaltest_functionalities/uart_send_char
test_functionalities/wifi_jpeg_streamer
update stream examples to work with gapsdk 3.6 and up聽#19 still remainsI looked into the code a little bit and it looks like some sort of race condition to me.When I insert aprintf
-statement before this line: https://github.com/GreenWaves-Technologies/gap_sdk/blob/0e4b0426510201906a3c9db4997e71dc78aeca8a/tools/gap_tools/frame_streamer/frame_streamer.c#L219 the blocking is gone, but streaming is obviously very slow (less than 1FPS according to the python-viewer)image_processing_examples/FaceDetection
working correctly.accurate sometimesBut since similar towifi_jpeg_streamer
the streaming is also not working, it's hard to debugimage_processing_examples/image_manipulations
ai_examples/gapflow_model_extern
mnist
-->model
tensorflow
from therequirements.txt
-list for thenntool
I manually added in the docker installationai_examples/mnist_simple
Seems completely brokenIs also not updated in the official SDK-exampleAll examples were tested in the docker.
Happy to take some feedback. If I find time in the next few days I will probably investigate some of the issues that still remain.