applied-deep-learning-lab / yolov5_rk3588 Goto Github PK

Is it possible to launch this project with its own additionally trained and converted model without a global rework of the project? Are there any restrictions on the model in this case? Thanks in advance for your answer.

Where are the inference.py documents?

I have exported a yolov5s model to onnx and then converted it to rknn w/ This code from rknn

I have tried to edit the config file to work with this model.

Here is my config file.

{
    "debug": {
        "print_camera_release": false,
        "camera_release_file": "camera_release.txt",
        "showed_frame_id": false,
        "showed_id_file": "showed_frames.txt",
        "filled_frame_id": false,
        "filled_id_file": "filled_ids.txt",
        "verbose": false,
        "verbose_file": "verbose.txt"
    },
    "camera": {
        "show": false,
        "source": 0,
        "pixel_format": "MJPG",
        "width": 640,
        "height": 640,
        "fps": 60
    },
    "inference": {
        "net_size": 640,
        "buf_size": 3,
        "obj_thresh": 0.25,
        "nms_thresh": 0.45,
        "inf_proc": 3,
        "post_proc": 3,
        "async_mode": true,
        "default_model": "yolov5.rknn",
        "new_model": "yolov5.rknn",
        "classes": [
            "note"
 ]
    },
    "storages": {
        "state": true,
        "stored_data_amount": 300,
        "dets_amount": 100,
        "frames_delay": 10
    },
    "webui": {
        "state": false,
        "send_data_amount": 10
    },
    "bytetrack": {
        "state": false,
        "fps": 60,
        "tracking_classes": [
            1
        ]
    },
    "pulse_counter": {
        "state": true
    }
}

The model only has one class "note". Please help.

I am getting the error,

ctypes.ArgumentError: argument1: <class 'TypeError'>: wrong type

running on an orangepi5. followed directions on readme.

你好，你们这个模型是你基于官网的模型进行训练的吗

Hello
can you give some info about the fps of your Inference on this board?
i really appreciate it

Hello!
Where I can find inference.py file for Python3 inference?
And will it work for rk3588s?
Thank you!

Hello, I'm trying your yolov5s_leaky_352x352.rknn model.

1. Which colour space do you use (BGR/RGB)?
2. Input shape is (BS, H, W, C)?
3. How much classes in output? Which dataset did you use?
4. I tried to infer (1, 352, 352, 3) pic and got 3 anchors: (1, 255, 44, 44), (1, 255, 22, 22), (1, 255, 11, 11). Can you please explain me output? (where is classes, confidences, bb coordinates, etc.)

Hi! There are three models in your list of .rknn models, and by default you use yolov5m_leaky_352x352.rknn . I have several questions:

1. How did you get this .rknn model (and in general all the listed models)?

2. If you change the model in config.json to yolov5m_leaky, yolov5s-640-640, the program starts to work worse. What other variable values need to be changed in config.json and in main.py for the program to work as well as with yolov5m_leaky_352x352.rknn?

3. If I change the model in config.json to my custom model, then what needs to be changed in config.json and in main.py for detection to work?

Why my rknn.inference resultis (1,25200,85), while the official reasoning results are (1,255,80,80), (1,255,40,40), (1,255,20,20)