Maximum number of epochs or steps about hdrnet HOT 5 OPEN

The author uses tf.train.Supervisor for the training process, so the process will stop until an exception is raised or any other thread calls request_stop by some reason. You could setup the stop criteria by yourself, or simply press Ctrl+C to stop the whole process, then the process will save the checkpoint before it leaves.

from hdrnet.

Where do we define, or even infer, the maximum number of epochs or steps on the default model? I have an example running for almost 5 days with only 2000 images and I don't know when will it stop.

@rmozart , can you elaborate about the compilation process? Many people are struggling with it, and you made it.

What cuda & cudnn versions?
What TF version?
What std=c++?
What code version did you use (which commit?)
Do you have a docker for it? Can you share the .so file?

Thanks!

from hdrnet.

Where do we define, or even infer, the maximum number of epochs or steps on the default model? I have an example running for almost 5 days with only 2000 images and I don't know when will it stop.

@rmozart , can you elaborate about the compilation process? Many people are struggling with it, and you made it.

What cuda & cudnn versions? What TF version? What std=c++? What code version did you use (which commit?) Do you have a docker for it? Can you share the .so file?

Thanks!

Hi, I could also share some information for my compilation.

I compile the bilateral_slice.cu.cc with the environment:
TF2.5 source code, CUDA-10.0, std=c++11.
Unfortunately, that’s all I remember.

I have a note in Chinese to remind me how to solve the compilation errors. You could share where you are now, perhaps I could give you a hand:)

from hdrnet.

Where do we define, or even infer, the maximum number of epochs or steps on the default model? I have an example running for almost 5 days with only 2000 images and I don't know when will it stop.

@rmozart , can you elaborate about the compilation process? Many people are struggling with it, and you made it.
What cuda & cudnn versions? What TF version? What std=c++? What code version did you use (which commit?) Do you have a docker for it? Can you share the .so file?
Thanks!

Hi, I could also share some information for my compilation.

I compile the bilateral_slice.cu.cc with the environment: TF2.5 source code, CUDA-10.0, std=c++11. Unfortunately, that’s all I remember.

I have a note in Chinese to remind me how to solve the compilation errors. You could share where you are now, perhaps I could give you a hand:)

I forgot which commit I used before, but I remember the time I built it is around 2018. So I think you could try this commit:
5ac95ef

from hdrnet.

Where do we define, or even infer, the maximum number of epochs or steps on the default model? I have an example running for almost 5 days with only 2000 images and I don't know when will it stop.

@rmozart , can you elaborate about the compilation process? Many people are struggling with it, and you made it.
What cuda & cudnn versions? What TF version? What std=c++? What code version did you use (which commit?) Do you have a docker for it? Can you share the .so file?
Thanks!

Hi, I could also share some information for my compilation.
I compile the bilateral_slice.cu.cc with the environment: TF2.5 source code, CUDA-10.0, std=c++11. Unfortunately, that’s all I remember.
I have a note in Chinese to remind me how to solve the compilation errors. You could share where you are now, perhaps I could give you a hand:)

I forgot which commit I used before, but I remember the time I built it is around 2018. So I think you could try this commit: 5ac95ef

Thanks!
If you just could share your makefile it would be great

from hdrnet.