fine-tune settings

[hengrui0516]

Thanks for your work!
I wonder your finetune settings on doclaynet.
And I find that "The models were fine-tuned using 4xA100 GPUs on the Doclaynet-base dataset, which consists of 6910 training images, 648 validation images, and 499 test images." in readme.md. As far as I know, Doclaynet dataset is ten times bigger than those, containing about 69103 training images, 6480 validation images, and 4994 test images.

[moured]

Hello @hengrui0516 , thanks for the valuable comment. You're right; I missed the last digit when writing the README. The models were definitely trained on the whole dataset. I have updated* the correct values.

[sunzx8]

Hello @hengrui0516 , thanks for the valuable comment. You're right; I missed the last digit when writing the README. The models were definitely trained on the whole dataset. I will upload the correct values right away.

hi，could you tell me the hypermeters during training like epoch, imgsize,batch size? thank you!

[moured]

sure @sunzx8,

epochs: 100
batch: 64
imgsz: 640

[aalhayali]

Hi! What were the labels used for this model? Is there a repo where I can play around the code further than inference?

EDIT: I assume it's based on the labels provided in the dataset you pre-trained YOLO on. The labels are:

[moured]

@aalhayali exactly, I used the 11 classes which are provided with the DocLayNet dataset. If you wanna further modify and edit the model/training setting please refer to the Yolo-v10 repo : https://github.com/THU-MIG/yolov10