When I train my custom key-point model, I get the following warnings notifying me that my annotation format is incorrect:
train: WARNING: Ignoring corrupted image and/or label dataset\images\train\Vid3_064.jpg: cannot reshape array of size 33 into shape (2)
train: WARNING: Ignoring corrupted image and/or label dataset\images\train\Vid3_065.jpg: non-normalized or out of bounds coordinate labels
I then get this error:
ValueError: zero-size array to reduction operation maximum which has no identity
It's important to note that my model is not for pose estimation of an human body. It's a single class dataset.
I'm trying to train my custom model using YOLOv7-pose with:
python .\train.py --data .\data.yaml --epochs 10 --batch-size 8 --img-size 640 --weights .\yolov7-w6-pose.pt
I followed How to prepare a custom keypoints dataset for WongKinYiu/Yolov7 Pose Estimation? to setup my annotation file format. The format I followed:
[class_id][x_center][y_center][width][height][kpt1_x][kpt1_y][visibility_1][kpt2_x][kpt2_y][visibility_2]...[kptn_x][kptn_y][visibility_n]
Below is an example of an image label containing four instances of the single class object:
0 213.69 320.73 94.43 211.67 166.71 225.7 2 254.53 215.36 2 261.14 427.03 2 172.38 414.82 2 0 321.11 322.67 117.18 241.32 262.52 215.03 2 374.15 203.13 2 379.7 444.45 2 268.07 428.07 2 0 463.31 325.33 154.74 281.55 385.56 201.99 2 535.81 185.76 2 540.3 467.31 2 391.57 446.26 2 0 658.59 328.66 211.84 343.47 551.87 184.12 2 763.71 158.78 2 763.69 502.25 2 555.09 469.49 2
