dnntoolbox / net-trim

The latest version of Net-Trim which solves the actual constrained problem

Net-Trim

This page contains the latest version of Net-Trim, which addresses the original convex program in the constrained form. For simple instructions on how to use the code, please visit:

• Latest Code Manual
• Latest (2020) Journal Paper

To access the Net-Trim code in the regularized form (old version), please visit:

• Old Github Page
• NIPS (2017) Conference Paper

net-trim's Issues

Error when run ConvNetTrimSolver_tf.py

I ran the code of first part of II.1. Retraining the Convolutional Layers
(https://dnntoolbox.github.io/Net-Trim/#ii-net-trim-code-step-by-step-demo)
Error occurs in the line:
W_nt = nt_conv.run(x, y, V, b, W.shape, epsilon, rho=2, num_iterations=max_iterations)

Error message:
InvalidArgumentError: filter must be 4-dimensional: [128]
[[Node: Conv2D = Conv2D[T=DT_FLOAT, data_format="NHWC", padding="SAME", strides=[1, 1, 1, 1], use_cudnn_on_gpu=true, _device="/job:localhost/replica:0/task:0/device:CPU:0"](_arg_Placeholder_0_0, _arg_Placeholder_7_0_7)]]

And I checked the arguments:
x shape = (20, 60, 60, 256)
y shape = (20, 60, 60, 256)
V shape = (20, 60, 60, 256)
b shape = (20, 60, 60, 256)
W shape = (20, 60, 60, 256)

I have no idea what wrong is.
Appreciate your help!