Yolov4 training

[nayamac]

@AlexeyAB I am using yolov4, my map is vey low so I used -show_imgs flag at the end of training command, but most of the images Were cropped and there was no bounding boxes in them, My question is why were the images cropped ?

[AlexeyAB]

@nayamac

I am using yolov4, my map is vey low

Show screenshot of your mAP.

---

but most of the images Were cropped

It is normal. It is just data augmentation.

---

there was no bounding boxes in them

Labels in your dataset are wrong. Check them by using one of these utils: https://github.com/AlexeyAB/darknet#how-to-mark-bounded-boxes-of-objects-and-create-annotation-files

[AlexeyAB]

Try to train 6000 iterations.

[nayamac]

When i tried 6000 iterations the results were 9%

[AlexeyAB]

-show_imgs flag at the end of training command, but most of the images Were cropped and there was no bounding boxes in them,

If you don't see bboxes, then your labels are wrong. Try to put txt-files with labels into the same directory where are images. And re-check labels. Check that you saved them in yolo-format.

[nayamac]

I saved them in yolo format, the txt-files are in the same directory with the images, that's why i don't know what went wrong, i followed every step, but the results didn't get any better

[AlexeyAB]

1. Can you show several screenshots of images from -show_imgs ?
2. Can you attach the training log?

[stephanecharette]

Load your images and annotations in DarkMark. My money is the annotations aren't right, and DarkMark will immediately show you if that is the case.

[nayamac]

Okey i will