Releases: Applied-Deep-Learning-Lab/Yolov5_RK3588
Restructuring project, new web graphical user interface
New project structure
- The startup file is now in a prominent place and you don't have to search for it in folders.
- Also , the base program files and add-on files are separated into folders.
- In addition, there is a configuration file, where you can change different parameters, enable and disable add-ons.
New web graphical user interface
- The web user interface has been redesigned.
- Now you can not only download new models, but also choose from those already downloaded.
- The settings can be changed by loading a configuration file.
- Also switching between the video stream and settings is implemented through buttons (links) in the GUI.
- In addition, you can apply settings changes, restart the program, or reboot the device directly from the GUI.
v0.3 Restructuring project for your own addons
Run your own addon with inference in parallel
In this release, we have separated the entire recognition process into a separate file.
Now you can write your own code in parallel with recognition and take data from it for yourself without adding the class itself.
And this is done in the "main.py" file, in which the class "rk3588" is imported, an object is created and all the recognition processes are started via the method "start".
v0.2.1 Aiortc server, change camera settings and models while running
Added aiortc server
With this release, you can watch the results of the inference from anywhere if you are connected to your device via LAN. Just type the local IP of the device and port "8080" in the address bar of your browser, and then click ''start inference''.
-
Model changing
To change the model, click ''Choose model file...'', then specify the model path and click ''Upload model''.
-
Camera settings changing
To change the camera settings add "/settings" to the address bar, then change whatever you need (for the camera) and click "Update settings".
v0.2 ByteTracker, Storages for DB, Inference in ONE FILE
This release incorporates:
-
Common inference (main.py)
Showing frames with common bounding box:
-
Inference with ByteTrack (main_bytetracker.py)
Showing frames with ids for selected classes
-
Inference with ByteTrack and Storages (main_database.py)
Showing frames with ids for selected classes and storage raw frames, inferenced frames, detections (bounding boxes, classes, scores/classes ids) in arrays (which use shared memory).
v0.1 Multiprocessing realtime Yolov5 on RK3588
Realtime inference using Yolov5 on RK3588's NPU. All 3 NPU cores are used.
Camera reading(+ resizing frames), inference, post processing, frames showing used in different processes. For inference and post processing you can specify the number of processes. Data is transferred between processes via queues.