Train custom dataset

[wongkinyiu]

Train custom dataset

[WongKinYiu]

https://github.com/WongKinYiu/yolov7#training

[yeyan00]

train demo

dataset: coco128,

set coco.yaml

train: images/train2017  # train images (relative to 'path') 128 images  
val: images/train2017  # val images (relative to 'path') 128 images

set train.py

    parser.add_argument('--weights', type=str, default='runs/yolov7.pt', help='initial weights path')  
    parser.add_argument('--cfg', type=str, default='cfg/training/yolov7.yaml', help='model.yaml path')  
    parser.add_argument('--data', type=str, default='data/coco.yaml', help='data.yaml path')  
    parser.add_argument('--hyp', type=str, default='data/hyp.scratch.p5.yaml', help='hyperparameters path')

convert your dataset like coco128 or coco, either modify util/dataset.py read your dataset

[thongvhoang]

train demo

dataset: coco128,

set coco.yaml

train: images/train2017  # train images (relative to 'path') 128 images  
val: images/train2017  # val images (relative to 'path') 128 images

set train.py

    parser.add_argument('--weights', type=str, default='runs/yolov7.pt', help='initial weights path')  
    parser.add_argument('--cfg', type=str, default='cfg/training/yolov7.yaml', help='model.yaml path')  
    parser.add_argument('--data', type=str, default='data/coco.yaml', help='data.yaml path')  
    parser.add_argument('--hyp', type=str, default='data/hyp.scratch.p5.yaml', help='hyperparameters path')

convert your dataset like coco128 or coco, either modify util/dataset.py read your dataset

As I understand, the custom dataset format like YOLOv5, you can download the dataset https://github.com/ultralytics/yolov5/releases/download/v1.0/coco2017labels.zip to understand the dataset's structure. This link in the script [yolov7](https://github.com/WongKinYiu/yolov7)/[scripts](https://github.com/WongKinYiu/yolov7/tree/main/scripts)/get_coco.sh

[zye1996]

https://github.com/WongKinYiu/yolov7#training

The training fails when set --rect flag. It seems that the labels/segments array length do not match in datasets when doing paste_in augmentation.

[Zagreus98]

https://github.com/WongKinYiu/yolov7#training

The training fails when set --rect flag. It seems that the labels/segments array length do not match in datasets when doing paste_in augmentation.

Did you manage to solve the problem? I get recall 0 when training with --rect flag.