Has anyone successfully trained their own data?

[Rhroan]

I try v9-s v9-m v9-c rd-9c
used 100 images, just want find sticker, but I can't train a model

python yolo/lazy.py task=train name=test task.data.batch_size=16 model=rd-9c task.epoch=100 device=0 out_path="d:\temp\yolo"

[Rhroan]

https://drive.google.com/file/d/1ZjAGnYeLOWjGJ804qs0a51gsYtNDVzf1/view?usp=drive_link

[Rhroan]

I think I got answer

My dataset only 1 class , when enable Mosaic: 1
It no used and get more trouble

[darinhitchings]

You can't train anything with 100 training images. You need more than 10000 to get started. You need to find a labelled dataset such as COCO 2017 or something else and train on that to get started. Then do 'transfer learning' after the training convergences 70-80% of the way on the original data set.