search

Noba1anc3 / Publaynet

10 stars 3 forks source link

Steps of finetuning on my own dataset #2

Open Howardqlz opened 3 years ago

Howardqlz commented 3 years ago

Hello, thanks for your detail code. But i still a little confuse about the steps of finetuning on my own dataset.For example, if i want to use the model which already pretrained on the Publaynet as a pretrained model, then finetuning on my own dataset

Howardqlz commented 3 years ago

And the train-val-test dataset link、model pretrained on Publaynet dataset are all Invalidation.

Howardqlz commented 3 years ago

Or maybe you can give a example about the data format under "./data/train/images/" and "./data/train/train.json" that can let us modify our data

Noba1anc3 commented 3 years ago

Or maybe you can give a example about the data format under "./data/train/images/" and "./data/train/train.json" that can let us modify our data

Hi Howardqlz, if you want to find a example about the data format about images and json file, you can refer to the README file, there are urls of the example data.

Under the node of Download in README file, there are images and annotation, you can check it out.

Howardqlz commented 3 years ago

Or maybe you can give a example about the data format under "./data/train/images/" and "./data/train/train.json" that can let us modify our data

Hi Howardqlz, if you want to find a example about the data format about images and json file, you can refer to the README file, there are urls of the example data.

Under the node of Download in README file, there are images and annotation, you can check it out.

Thanks for your answer. Yes i check the ALL FILES under the Download in README file. But the link of Dataset under the Download are all invalidation. But i think i just need to segment the images and json file.And another quite important thing is that the link of model pretrained on the Publaynet is also invalidation. Because it's time-consuming to trian again, i will really appreciate if you can update the link again! ^ .^