search

dme-compunet / YoloV8

Integrate YOLOv8 into your C# project for a variety of real-time tasks including object detection, instance segmentation, pose estimation and more, using ONNX Runtime.
https://www.nuget.org/packages/YoloV8
GNU Affero General Public License v3.0
242 stars 45 forks source link

DetectObb is not showing any bounding boxes for object (car in my case) #53

Closed aloksharma1 closed 3 months ago

aloksharma1 commented 4 months ago

Hello i tried to use yolov8x-obb.onnx with your package because i need the object orientation as well as detection, but even on lowest confidence level i am not getting any detection. can you share an example for obb? here is my code, hopefully i am following the process correctly:

private readonly YoloV8Predictor _predictor= YoloV8Predictor.Create(@"G:\Dotnet\Desktop\Windows\ATTSystem\ATTSystem\Assets\yolov8x-obb.onnx");

using var image = Image.Load<Rgb48>(frame.ToBytes());
var result = _predictor.DetectObb(image); //no bounding boxes tried car /bus/truck/van images
dme-compunet commented 4 months ago

odd.... I will check it on my side soon.

dme-compunet commented 4 months ago

Can you attach the picture to reproduce the problem?

Message ID: @.***>

aloksharma1 commented 4 months ago

Hi i am using video with opencv to extract frame by frame images, i have extracted the images from my video example for your testing. you can use these to see why i am not getting anything on obb, fyi these same images work with segmentation and detect algo. what i need now is to not only segment the vehicle but also see its direction. car-parking-1_1 car-parking-1_2 car-parking-1_3 car-parking-1_4

dme-compunet commented 4 months ago

@aloksharma1 you will probably need to train your own model for your need, because the pretrained model is suitable for satellite photos and such (DOTAv1 dataset), not for low-altitude photos.

aloksharma1 commented 4 months ago

@aloksharma1 you will probably need to train your own model for your need, because the pretrained model is suitable for satellite photos and such (DOTAv1 dataset), not for low-altitude photos.

oh understood, then my next issue is i need to mix segmentation technique with oob what do you suggest a single custom trained model or using 2 separate models(one for obb and one for segmentation)?

dme-compunet commented 4 months ago

I do not know. You should ask here: https://github.com/ultralytics/ultralytics