Hi @LulinZhang,
According to the error, the training is running on your CPU instead of the GPU: "InvalidArgumentError: Default MaxPoolingOp only supports NHWC on device type CPU". Could you double-check that your configuration is correct and that the GPU is correctly recognized and used during the training?
Also, you get a lot of deprecation messages, which is surprising if you are running with TF 1.14. It seems as if you were using TF 2 already... Not sure why you get so many warnings, because it shouldn't happen with TF 1.14.
Hi @LulinZhang, According to the error, the training is running on your CPU instead of the GPU: "InvalidArgumentError: Default MaxPoolingOp only supports NHWC on device type CPU". Could you double-check that your configuration is correct and that the GPU is correctly recognized and used during the training?
Also, you get a lot of deprecation messages, which is surprising if you are running with TF 1.14. It seems as if you were using TF 2 already... Not sure why you get so many warnings, because it shouldn't happen with TF 1.14.
Originally posted by @rpautrat in https://github.com/rpautrat/SuperPoint/issues/185#issuecomment-750840579