ying09 / TextFuseNet

A PyTorch implementation of "TextFuseNet: Scene Text Detection with Richer Fused Features".
MIT License
476 stars 123 forks source link

Testing on custom datasets #73

Open Gokulnath-Sourirajan opened 3 years ago

Gokulnath-Sourirajan commented 3 years ago

Hi, after having downloaded the pretrained model, is there a way to test them on my custom dataset or ant other datasets that do not have its corresponding .py file in demo folder? Thanks!

Aditya-5895 commented 3 years ago

put images inside input_images folder give the path of downloded model in demo/*det.py and run demo/det.py. output will be saved in test_icdar** file

ipheiman commented 3 years ago

Hi there, I have followed Aditya's instructions. But when i run "python demo/icdar2015_detection.py", it takes quite awhile (10-15 minutes) before generating an error "RuntimeError: cuDNN error: CUDNN_STATUS_MAPPING_ERROR".

I thought it could be due to the batch size? So I changed the IMS_PER_BATCH in configs yml from 4 to 1, but the same error popped up. Have you encountered this?

Aditya-5895 commented 3 years ago

I encountered That error but after reducing batch size it worked. Its a cuda bug .If IMS_PER_BATCH=1 is also not working for you check without cuda or check proper pytorch or cuda version for your GPU. check with this link https://github.com/pytorch/pytorch/issues/27588

ipheiman commented 3 years ago

Thanks Aditya for the quick reply. You're probably right that its a cuda bug, and my GPU is RTX3080 which could be incompatible with cuda 10. Cheers!

kaxapatel commented 2 years ago

Can I use SVHN dataset on this model?