I am working on a project and I have the aim of learning how to use a Unet. Hence, I found this project (https://github.com/hlamba28/UNET-TGS/blob/master/TGS%20UNET.ipynb) and decided to adapt the code, but it mostly ressembles this, because I wanted to test it. However, as an added value, I wanted to use my GPU for the training not to last forever. I installed Cuda, Cudnn, added the environement variables on Windows, and installed tensorflow-gpu.
For an unknown reason, the kernel always dies when I fit the model at the first epoch, maybe 10 seconds after starting the fit. I checked the gpu information, and witnessed that the VRAM is filled to the max, and I thought it came from that. However, I limited the amount of VRAM that Python could use and it still crashed. I may have badly installed CUda and/or CudNN but I don't know how to check that. Has someone have any idea what I did wrong ?