Raspberry Pi - Submissions for EXPORT competition #3430
Replies: 10 comments 51 replies
-
|
https://github.com/karanjakhar/yolov5-export-to-raspberry-pi Please have a look. |
Beta Was this translation helpful? Give feedback.
-
|
@karanjakhar did you benchmark your solution in raspi (or any other device)? If yes then what FPS did you get? |
Beta Was this translation helpful? Give feedback.
-
|
Hi @karanjakhar for inferencing on video, the code is loading the tflite model for each frame? That's why it is giving slow results as well. |
Beta Was this translation helpful? Give feedback.
-
|
@rajchem You are right. Thank you for pointing it out. I have updated it. |
Beta Was this translation helpful? Give feedback.
-
|
@karanjakhar can you help how can I adjust the code for INT8 quantized tflite model? |
Beta Was this translation helpful? Give feedback.
-
|
@karanjakhar |
Beta Was this translation helpful? Give feedback.
-
|
Hello, I've create my repository to export YOLOv5 model from Pytorch framework to MNN framework (MNN is a lightweight deep neural network inference engine, embedded devices, such as IoT). Please take a look, feel free to comment or create an issue. Hope you enjoy it. |
Beta Was this translation helpful? Give feedback.
-
|
Thanks. We will look at it |
Beta Was this translation helpful? Give feedback.
-
|
Good Job. |
Beta Was this translation helpful? Give feedback.
-
|
@nocolour I've not compared my model with tflite, I think you did. But my code default saves image/video/webcam into local. Maybe it makes my code slower than your tflite model? Or you had add --nodisplay --nosave when inference? |
Beta Was this translation helpful? Give feedback.
-
|
@AnhPC03 good job man! btw, what FPS did you get (on the colab cpu and/or raspi)? |
Beta Was this translation helpful? Give feedback.
-
|
Why there is a need to export Pytorch model to other for making it run on Raspberry Pi? |
Beta Was this translation helpful? Give feedback.
-
|
Does anyone have any idea about this issue, please? - #4167 Would highly appreciate it if someone can help in know this. |
Beta Was this translation helpful? Give feedback.
-
|
@batrlatom @glenn-jocher I just have a key question that what are we trying to do with exportation of current Yolov5 into the exported version? What is the need to export for Raspberry Pi like devices? What is the difference between current yolov5 and the exported version of yolov5? Can anyone of you please let me know? I would highly appreciate that. |
Beta Was this translation helpful? Give feedback.
-
|
@jaskiratsingh2000 Hi, the goal is to make a yolov5 inference as fast as possible on Raspi. In normal situation, yolov5 is not quite realtime on such system, so applying quantization or any other technique which makes it faster is what it needs to be done. |
Beta Was this translation helpful? Give feedback.
-
|
@batrlatom Can you be more specific please because I am a researcher working with Yolo and Raspberry Pi and this is something very important for me to understand in terms of more detailed specifications. Thanks and Regards |
Beta Was this translation helpful? Give feedback.
-
|
@jaskiratsingh2000 I do not own raspi now, so I am unable to exactly try out the installation. But are you sure you have everything the same? The same checkout version, all libraries are on the same version? Computers are the same, but software layers could be different. If you do the |
Beta Was this translation helpful? Give feedback.
-
|
@batrlatom Yes. Every Library on both the machines is the same. I tried the implementation of the others but none of them actually have implemented the Accuracy Performance and due to this reason only I was not able to understand the key question that what is the difference between the current Yolo version and the Exported version. |
Beta Was this translation helpful? Give feedback.
-
|
in utils/general.py is a function |
Beta Was this translation helpful? Give feedback.
-
|
@karanjakhar I am still looking forward to hearing your response on this - https://docs.google.com/document/d/1ej8hIktTFmsqTH0GCYVw-qD8sF9C9fa_Pe61EenNmkU/edit?usp=sharing Can you please let me know? If anyone else also has any idea about this then please let me know since I and @batrlatom have been discussing this but couldn't reach a conclusion. Your response would be highly appreciable. Thanks! |
Beta Was this translation helpful? Give feedback.
-
|
@jaskiratsingh2000 Please solve the problem in the thread which is separate from the competition itself, thank you |
Beta Was this translation helpful? Give feedback.
-
|
Hi, I am a newbie, do we need to test it on Raspberry Pi device. |
Beta Was this translation helpful? Give feedback.
-
|
Hello, the new realease of yolo bring a amazing export model witch work perfectly on raspberry pi. With a 320 image, it is possible to reach 3 FPS and works perfectly with the raspberry camera. |
Beta Was this translation helpful? Give feedback.
-
|
Okey, that is what I was thinking.... Thanks for sharing! 👍🏼 |
Beta Was this translation helpful? Give feedback.
-
|
One question regarding execution times. The inference time using the demo images is around 450 ms using int8 tflite model. Can someone give some advice in how to optimize this times and further steps? Thanks a lot for the help of this community! 👍🏼 |
Beta Was this translation helpful? Give feedback.
-
Hello, the image is available here : https://drive.google.com/file/d/1WrhaWJL2_eIJKJHgdf5KFnYqruFVlq2R/view?usp=sharing |
Beta Was this translation helpful? Give feedback.
-
|
Nice! Thank you @blange48 |
Beta Was this translation helpful? Give feedback.
-
|
@blange48 I just have a quick question to understand regarding which I am having a hard time understanding it since long. What is actually inside the Pytorch model and Tflite model? I don't understand that what's the format inside the PyTorch model. When do we say that we are compiling the Pytorh model then what does happen intermediately? Also, @blange48 Were you able to get the mAP value of any model on the RPi. I am tagging other people if in case they have any idea about this then that would really lead to a healthy discussion. |
Beta Was this translation helpful? Give feedback.
-
|
Hey Everyone! @glenn-jocher @zldrobit @blange48 I just have a question that what is the core difference between the Pytorch Model and the Tflite model? As I could see that Pytorch model is not working good on the Raspberry Pi whereas the Tflite works better but I am not able to figure out at what are the reasons for that its working. |
Beta Was this translation helpful? Give feedback.
-
|
Hi participants, Thank you all for your submissions, we have already reviewed them and are about to announce the winners! Please send me an email at stefani@ultralytics.com including your full name and GitHub username, so I can share more information with you. Stay tuned - the results are coming soon! Stefani Kovachevska |
Beta Was this translation helpful? Give feedback.
-
|
Hi everyone, We were thrilled to have received your submissions to our YOLOv5 Export Competition for the Raspberry Pi category. Unfortunately, none of the submissions passed the minimum set of requirements in each of the evaluation criteria. Therefore, no winner was selected for the category this time. However, there will be more chances for participants to compete again in the future. Thank you all for your efforts and for being a valuable member of our open-source community. Contributions from you are what makes our community great. 🙌 Be sure to check out the winning repositories in the rest of the categories in our latest blog post: https://ultralytics.com/article/export-competition-winners Stay amazing and keep creating! 🚀 |
Beta Was this translation helpful? Give feedback.
-
Use this discussion thread for YOLOv5 🚀 EXPORT Competition submissions in the Raspberry Pi category. Good luck!
Beta Was this translation helpful? Give feedback.
All reactions