core\framework\op_kernel.cc:1192] Resource exhausted: OOM when allocating tensor with shape[64,256,16,16]

[c1a1o1]

core\framework\op_kernel.cc:1192] Resource exhausted: OOM when allocating tensor with shape[64,256,16,16]

When I train it??

[Prinsphield]

Please give more details about your experiment environment. Maybe you can scale down the batch size.

[c1a1o1]

Caused by op 'decoder_6/deconv2/conv2d_transpose', defined at: File "D:/work/gansgood/DNA-GAN-master/train.py", line 133, in main() File "D:/work/gansgood/DNA-GAN-master/train.py", line 128, in main DNA_GAN = Model(args.attributes, is_train=True) File "D:\work\gansgood\DNA-GAN-master\model.py", line 26, in init self.build_model() File "D:\work\gansgood\DNA-GAN-master\model.py", line 224, in build_model self.Bxs = [self.joiner('decoder', enc_B) for enc_B in self.enc_Bxs] File "D:\work\gansgood\DNA-GAN-master\model.py", line 224, in self.Bxs = [self.joiner('decoder', enc_B) for enc_B in self.enc_Bxs] File "D:\work\gansgood\DNA-GAN-master\model.py", line 174, in joiner strides=[1,2,2,1]) File "D:\work\gansgood\DNA-GAN-master\model.py", line 94, in make_deconv_bn output_shape=out_shape, strides=strides, padding='SAME'

[c1a1o1]

gtx1070

[Prinsphield]

I have tested it in different linux machines but no windows machine. I sorry for being not able to reproduce that. Maybe you can modify the dataset.py and change the batch size to a smaller number (e.g., 16) and start to train with respect to no more than 2 attributes.