Darknet: YOLOv2 training gives "Segmentation fault" after some iterations

[xhuvom]

I am trying to train a custom dataset using YOLOv2 on a GTX1080 card with Batch size = 64 and Subdivisions = 8. With the tiny-yolo (& any .cfg) after some iterations ( 10~100), it gives segmentation fault (core dumped). I am sure to put the images and labels properly formatted. I have tried with OPENCV = 0 but nothing improved. I also cant figure out how to use the DEBUG option.
I have tried with different Batch and Subdivision numbers but in vein. I had trained earlier version of YOLO easily but the new version is giving me headache.

Can anyone help?

[xhuvom]

I can train COCO dataset without any error. However, my own dataset gives the Segmentation fault. I had triple checked the directory structures and label formats.

[xgl1991714]

maybe there are some invalid coordinate data like zero in your label files

[xhuvom]

Yes I had figured it out. Thanks although.

[gentlebreeze1]

how to solve coordinate data like zero in your label files？
@xhuvom