Change size

[williamd4112]

I want to predict 84x84 grey scale image. How to change the input size in code ?
I try replacing all 210x160x3 definition in the net.py but it doesn't work.

Thanks

The following is error message:

I0123 23:24:49.122573 2669 net.cpp:118] Creating Layer x_hat
I0123 23:24:49.122580 2669 net.cpp:492] x_hat <- x_hat-05
I0123 23:24:49.122586 2669 net.cpp:448] x_hat -> x_hat
I0123 23:24:49.122603 2669 net.cpp:163] Setting up x_hat
I0123 23:24:49.122611 2669 net.cpp:171] Top shape: 32 3 210 160 (3225600)
I0123 23:24:49.122617 2669 net.cpp:202] Num of parameters : 0
I0123 23:24:49.122620 2669 layer_factory.hpp:76] Creating layer silence
I0123 23:24:49.122634 2669 net.cpp:118] Creating Layer silence
I0123 23:24:49.122640 2669 net.cpp:492] silence <- act-01
I0123 23:24:49.122648 2669 net.cpp:492] silence <- act-02
I0123 23:24:49.122653 2669 net.cpp:492] silence <- act-03
I0123 23:24:49.122659 2669 net.cpp:163] Setting up silence
I0123 23:24:49.122663 2669 net.cpp:202] Num of parameters : 0
I0123 23:24:49.122668 2669 layer_factory.hpp:76] Creating layer l2_loss
I0123 23:24:49.122674 2669 net.cpp:118] Creating Layer l2_loss
I0123 23:24:49.122679 2669 net.cpp:492] l2_loss <- x_hat
I0123 23:24:49.122684 2669 net.cpp:492] l2_loss <- label
I0123 23:24:49.122692 2669 net.cpp:448] l2_loss -> l2_loss
F0123 23:24:49.122704 2669 euclidean_loss_layer.cpp:14] Check failed: bottom[0]->count(1) == bottom[1]->count(1) (100800 vs. 7056) Inputs must have the same dimension.

[junhyukoh]

Unfortunately, this repo does not include the network architecture used for 84x84 gray scale images.
You should modify net.py based on the details in the paper so that it can deal with 84x84 images.
Thanks.