Even though specify the correct class name and mapping data.yaml ,Model detecting and labeling random class

[thangarajdeivasikamani]

Hi, I have used the below dataset for object detection: https://www.kaggle.com/datasets/ashfakyeafi/road-vehicle-images-dataset

I have created dataset folder as below:

Data_1.yaml prepared as below:

Class: ['ambulance', 'army vehicle', 'auto rickshaw', 'bicycle', 'bus', 'car', 'garbagevan', 'human hauler', 'minibus', 'minivan', 'motorbike', 'pickup', 'policecar', 'rickshaw', 'scooter', 'suv', 'taxi', 'three wheelers -CNG-', 'truck', 'van', 'wheelbarrow']

Training command I have used :

training with single GPU support

!python tools/train.py --batch 32 --conf configs/yolov6s.py --epochs 200 --data /content/drive/MyDrive/YOLO_6_Custom/dataset/trafic_data/data_1.yaml --device 0

Around 75 Epocs completed:

After that I have done the inference as below:

!python tools/infer.py --weights /content/drive/MyDrive/YOLO_6_Custom/YOLOv6/runs/train/exp/weights/best_ckpt.pt --source /content/drive/MyDrive/YOLO_6_Custom/dataset/trafic_data/images/test --device 0

But I am getting wrong prediction labels:

[mtjhl]

Please check if the index of the labels is the same with your labels.

['ambulance', 'army vehicle', 'auto rickshaw', 'bicycle', 'bus', 'car', 'garbagevan', 'human hauler', 'minibus', 'minivan', 'motorbike', 'pickup', 'policecar',
'rickshaw', 'scooter', 'suv', 'taxi', 'three wheelers -CNG-', 'truck', 'van', 'wheelbarrow']

For example , army vehicle label's index is 1, is it 1 in the txt format labels?

[thangarajdeivasikamani]

Hi, The same data set and labeling working in YOLO V5.And also refer the attached detection screenshot. I am not sure whether the model get the labels examples Horse, Fire hydrant,etc

[mtjhl]

If YOLOv6 and v5 all have same problem, the class index maybe is wrong. You'd better visualize the index on image, to see if the index is wrong.

[danielchua2602]

try including --yaml /content/drive/MyDrive/YOLO_6_Custom/dataset/trafic_data/data_1.yaml when running your inference

[thangarajdeivasikamani]

@danielchua2602 Thanks now working