YOLOv8n is an object detection model. It uses bounding boxes to precisely draw each object in image.

Inference results on VIM4.

Inference speed test: USB camera about 190ms per frame. MIPI camera about 140ms per frame.

Download the YOLOv8 official code. ultralytics/ultralytics

$ git clone https://github.com/ultralytics/ultralytics

Refer README.md to create and train a YOLOv8n model. My version torch==1.10.1 and ultralytics==8.0.86.

Get source khadas/vim4_npu_sdk.

$ git lfs install
$ git lfs clone https://github.com/khadas/vim4_npu_sdk
$ cd vim4_npu_sdk
$ ls
adla-toolkit-binary  adla-toolkit-binary-3.1.7.4  convert-in-docker.sh  Dockerfile  docs  README.md

After training the model, modify ultralytics/ultralytics/nn/modules/head.py as follows.

head.py

diff --git a/ultralytics/nn/modules/head.py b/ultralytics/nn/modules/head.py
index 0b02eb3..0a6e43a 100644
--- a/ultralytics/nn/modules/head.py
+++ b/ultralytics/nn/modules/head.py
@@ -42,6 +42,9 @@ class Detect(nn.Module):
 
     def forward(self, x):
         """Concatenates and returns predicted bounding boxes and class probabilities."""
+        if torch.onnx.is_in_onnx_export():
+            return self.forward_export(x)
+
         shape = x[0].shape  # BCHW
         for i in range(self.nl):
             x[i] = torch.cat((self.cv2[i](x[i]), self.cv3[i](x[i])), 1)
@@ -80,6 +83,15 @@ class Detect(nn.Module):
             a[-1].bias.data[:] = 1.0  # box
             b[-1].bias.data[:m.nc] = math.log(5 / m.nc / (640 / s) ** 2)  # cls (.01 objects, 80 classes, 640 img)
 
+    def forward_export(self, x):
+        results = []
+        for i in range(self.nl):
+            dfl = self.cv2[i](x[i]).contiguous()
+            cls = self.cv3[i](x[i]).contiguous()
+            results.append(torch.cat([cls, dfl], 1))
+        return tuple(results)
+

Create a python file written as follows to export ONNX model.

export.py

from ultralytics import YOLO
model = YOLO("./runs/detect/train/weights/best.pt")
results = model.export(format="onnx")

$ python export.py

Pull yolov8n.onnx model into vim4_npu_sdk/adla-toolkit-binary-3.1.7.4/python and then run convert-in-docker.sh

$ ./convert-in-docker.sh ksnn

If your yolov8n model parameters are different from ours, you can change parameters in ksnn_args.txt.

Download KSNN library and demo code. khadas/ksnn-vim4

$ git clone https://github.com/khadas/ksnn-vim4.git

If you use Ubuntu 24.04, demo must run in python virtual environment.

Create and start python virtual environment.

$ sudo apt update
$ sudo apt install python3-venv
$ python3 -m venv myenv
$ source myenv/bin/activate

Install KSNN package.

$ cd ksnn-vim4/ksnn
$ sudo apt update
$ sudo apt install python3-pip
$ pip3 install ksnn_vim4-1.4.1-py3-none-any.whl

Put yolov8n.nb and libnn_yolov8n.so into ksnn/examples/yolov8n/models/VIM4 and ksnn/examples/yolov8n/libs

If your model's classes is not 80, please remember to modify the parameter, LISTSIZE.

LISTSIZE = classes number + 64

$ cd ksnn/examples/yolov8n
$ export QT_QPA_PLATFORM=xcb
$ python3 yolov8n-picture.py --model ./models/VIM4/yolov8n_int8.adla --library ./libs/libnn_yolov8n.so --picture ./data/horses.jpg --level 0

$ cd ksnn/examples/yolov8n
$ export QT_QPA_PLATFORM=xcb
 
# USB Camera
$ python3 yolov8n-cap.py --model ./models/VIM4/yolov8n_int8.adla --library ./libs/libnn_yolov8n.so --type usb --device 0
 
# MIPI Camera
$ python3 yolov8n-cap.py --model ./models/VIM4/yolov8n_int8.adla --library ./libs/libnn_yolov8n.so --type mipi --device 63

0 is the camera device index.

$ pip3 uninstall opencv-python
$ sudo apt install python3-opencv