search

GXYM / TextBPN-Plus-Plus

Arbitrary Shape Text Detection via Boundary Transformer；The paper at: https://arxiv.org/abs/2205.05320, which has been accepted by IEEE Transactions on Multimedia (T-MM 2023).
175 stars 38 forks source link

About datasetes download #6

Closed HHeracles closed 1 year ago

HHeracles commented 2 years ago

Thank you very much for this job。But the shared datasetes link on the Google drive cannot be opened, can you share it through the Baidu Drive?

GXYM commented 2 years ago

Thank you very much for this job。But the shared datasetes link on the Google drive cannot be opened, can you share it through the Baidu Drive?

You can download them from offical website: https://github.com/cs-chan/Total-Text-Dataset

HHeracles commented 2 years ago

Thanks for your reply！I downloaded the dataset myself from the official website and trained it. However, it is difficult to reproduce the performance in your paper.If you can provide us with your processed datasets and corresponding trained models, maybe we can reproduce the performance of the thesis theory, that would be great. If you can't provide the datasets, can you provide models trained on Total-Text, CTW-1500, MSRA-TD500, etc?Thanks again for jobs.

GXYM commented 2 years ago

Thanks for your reply！I downloaded the dataset myself from the official website and trained it. However, it is difficult to reproduce the performance in your paper.If you can provide us with your processed datasets and corresponding trained models, maybe we can reproduce the performance of the thesis theory, that would be great. If you can't provide the datasets, can you provide models trained on Total-Text, CTW-1500, MSRA-TD500, etc?Thanks again for jobs.

we just use the official datasets (old version) without processing, and we also have provided the trained model links in README, such as image

HHeracles commented 2 years ago

Thanks for you relpy. I have one more question. I downloaded MLT-2017 from the official website, but “data/MLT2017/train_list.txt” is displayed when running the training program. How to generate the “train_list.txt” from the MLT2017 of the official website.

GXYM commented 2 years ago

Thanks for you relpy. I have one more question. I downloaded MLT-2017 from the official website, but “data/MLT2017/train_list.txt” is displayed when running the training program. How to generate the “train_list.txt” from the MLT2017 of the official website.

You can use the cmd "ls mlt_train/*.jpg>> train_list.txt" , writing the img path to txt file, such as "mlt_train/xxxxx.jpg"

123cjjjj commented 1 year ago

Hello, have you successfully trained on MLT?