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 Parameter Settings for eval. #27

Open HHeracles opened 11 months ago

HHeracles commented 11 months ago

Thank you for contributing to the nice work. But there's one thing I don't quite understand, and I'm hoping to get your answer to it.

When verifying the model's performance on a dataset such as total-text, "tp" and "tr" use two different sets of values. What do "tp" and "tr "mean? If improvements are made based on this model, should these two values be kept the same for "tr" and "tp" when comparing performance?

GXYM commented 11 months ago

Thank you for contributing to the nice work. But there's one thing I don't quite understand, and I'm hoping to get your answer to it.

When verifying the model's performance on a dataset such as total-text, "tp" and "tr" use two different sets of values. What do "tp" and "tr "mean? If improvements are made based on this model, should these two values be kept the same for "tr" and "tp" when comparing performance?

please refer to https://github.com/cs-chan/Total-Text-Dataset. This is the official evaluation method, not something we can change.