search

jason-li-831202 / Vehicle-CV-ADAS

The project can achieve FCWS, LDWS, and LKAS functions solely using only visual sensors. using YOLOv5 / YOLOv5-lite / YOLOv6 / YOLOv7 / YOLOv8 / YOLOv9 / EfficientDet and Ultra-Fast-Lane-Detection-v2 .
GNU General Public License v3.0
162 stars 38 forks source link

Update for yolov10 support #23

Closed AekSW closed 2 months ago

AekSW commented 3 months ago

Is it possible to use yolov10? I tried using it but it broke the utils for Bytetrack.

Traceback (most recent call last):
  File "C:\Projects\Vehicle-CV-ADAS-master\demo.py", line 304, in <module>
    objectTracker.DrawTrackedOnFrame(frame_show, False)
  File "C:\Projects\Vehicle-CV-ADAS-master\ObjectTracker\byteTrack\byteTracker.py", line 213, in DrawTrackedOnFrame
    self.plot_trajectories(frame, trajector , cid, tid)
  File "C:\Projects\Vehicle-CV-ADAS-master\ObjectTracker\core.py", line 195, in plot_trajectories
    color=self.class_colors[class_id],
KeyError: 'unknown'

What can I do to fix it? Will there be any update for yolov10 support?

jason-li-831202 commented 3 months ago

YOLOv10 has added some post-processing parts compared to the previous YOLO series. Therefore, there are two ways to use it: either delete the original model's post-processing without converting it to ONNX, or write new post-processing for YOLOv10's output.

image