are you stiil here

[SabraHashemi]

please send me a reply if you are here question about pytorch craft i have same problem as you

[zcswdt]

please send me a reply if you are here question about pytorch craft i have same problem as you

what is the problem？

[SabraHashemi]

you have problem with time in craft_pytorch https://github.com/clovaai/CRAFT-pytorch/issues/45#issuecomment-555319176 you said that you located y[0,:,:,1].cpu().data.numpy() in test.py i located it too

also in different run of project average proccess time for each image get 2x sometimes

i tested it with pytorch1.4.0 and 1.6.0

i have 380*60 images and average process (each image) is 0.11s but in some of my runs it reached to 0.25

it confused me

did you solve the problem? do you have any idea?

i stuck here for so long time

[zcswdt]

did you solve the problem? do you have any idea?

i stuck here for so long time

Later, I checked the time-consuming of the model. It is indeed at this step, but it is limited by the pytorch framework. After the training is completed, if CUDA tensor wants to convert to Numpy data, it must first be converted to cpu tensor before it can be converted to Numpy data, so This step is still time-consuming. For other problems of the kind you mentioned, I think if you download according to the versions of the various libraries in the repo, there should be no problem. I suggest you check whether your environment is the same as that in the repo.

[SabraHashemi]

the requirement are so old for example pytorch is <1 i have downgraded from 1.6 to 1.4 but problem still exist

this is my differents runs

D:\CRAFT-pytorch-master>py test.py --trained_model=craft_mlt_25k.pth --test_folder=test Loading weights from checkpoint (craft_mlt_25k.pth) elapsed time : 11.8345365524292s

D:\CRAFT-pytorch-master>py test.py --trained_model=craft_mlt_25k.pth --test_folder=test Loading weights from checkpoint (craft_mlt_25k.pth) elapsed time : 12.25997543334961s

D:\CRAFT-pytorch-master>py test.py --trained_model=craft_mlt_25k.pth --test_folder=test Loading weights from checkpoint (craft_mlt_25k.pth) elapsed time : 13.054944276809692s

D:\CRAFT-pytorch-master>py test.py --trained_model=craft_mlt_25k.pth --test_folder=test Loading weights from checkpoint (craft_mlt_25k.pth) elapsed time : 12.536467552185059s

D:\CRAFT-pytorch-master>py test.py --trained_model=craft_mlt_25k.pth --test_folder=test Loading weights from checkpoint (craft_mlt_25k.pth) elapsed time : 13.861915826797485s

after a while after runninng multple time it reached to 5 s which is best

but i want to have 5s from second run at least (but now its so random)

[SabraHashemi]

after a while this is runtime D:\CRAFT-pytorch-master>py test.py --trained_model=craft_mlt_25k.pth --test_folder=test Loading weights from checkpoint (craft_mlt_25k.pth) elapsed time : 6.037157297134399s

[SabraHashemi]

and i dont think that requirement solve problem in this link he satisfied requirement but had problem https://github.com/clovaai/CRAFT-pytorch/issues/91#issue-556803876

[zcswdt]

and i dont think that requirement solve problem in this link he satisfied requirement but had problem https://github.com/clovaai/CRAFT-pytorch/issues/91#issue-556803876

I have not encountered a situation similar to yours. I tested using 800*800 pictures, and it took about 0.4s to detect a picture. However, the general model loading is a bit time-consuming during initialization. I suggest you to test more pictures to get the average value.