Hi there,
I’m trying to work with yolov8 pose in order to train a pose model to detect keypoints on a sports field. I have about 240 images. In my project I only see three options for exporting my data and those are COCO, YOLOv5 PyTorch, YOLOv8.
I use the YOLOV8 one in order to download my keypoint labelled data to train a pose model using this command:
!yolo task=pose mode=train model=yolov8x-pose.pt data={dataset.location}/data.yaml batch=50 epochs=500 imgsz=640 mosaic=0.0 plots=True
However the results I get are very odd. It is able to detect the field but the points are completely off. Here are the results
Note that it also stopped training early:
EarlyStopping: Training stopped early as no improvement observed in last 100 epochs. Best results observed at epoch 172, best model saved as best.pt.
To update EarlyStopping(patience=100) pass a new patience value, i.e. `patience=300` or use `patience=0` to disable EarlyStopping.
272 epochs completed in 0.335 hours.
Optimizer stripped from runs/pose/train2/weights/last.pt, 152.0MB
Optimizer stripped from runs/pose/train2/weights/best.pt, 152.0MB
Validating runs/pose/train2/weights/best.pt...
Ultralytics 8.3.165 🚀 Python-3.11.13 torch-2.6.0+cu124 CUDA:0 (NVIDIA A100-SXM4-40GB, 40507MiB)
YOLOv8x-pose summary (fused): 121 layers, 75,740,313 parameters, 0 gradients, 293.4 GFLOPs
Class Images Instances Box(P R mAP50 mAP50-95) Pose(P R mAP50 mAP50-95): 100% 1/1 [00:00<00:00, 2.32it/s]
all 46 46 0.999 1 0.995 0.993 0 0 0 0
Speed: 0.1ms preprocess, 4.2ms inference, 0.0ms loss, 1.1ms postprocess per image
Im thinking I either don’t have enough images or something is wrong with the format that it is not picking up the keypoints correctly because there looks to be little to no improvements for the keypoints. Is the format that I’m exporting to correct?
Any help would be appreciated.