salingo/pgdnn

[SIGGRAPH 2018] Predictive and Generative Deep Neural Networks for Object Functionality

pgdnn's Issues

iGEN-NET: ValueError: No variables to optimize.

When i trained iGEN-NET, it raised a ValueError:

Traceback (most recent call last):
File "train.py", line 91, in
optimizer_ts = tf.train.AdamOptimizer(learning_rate=LEARNING_RATE).minimize(iGEN.ts_loss, var_list=para_ts)
File "/usr/lib/python2.7/site-packages/tensorflow/python/training/optimizer.py", line 343, in minimize
grad_loss=grad_loss)
File "/usr/lib/python2.7/site-packages/tensorflow/python/training/optimizer.py", line 408, in compute_gradients
raise ValueError("No variables to optimize.")
ValueError: No variables to optimize.

I try to print para_ts and para_rec, but both of them are empty .

3D Spatial Transform Network

Hi,
I see you did the 3d transform in 2 step as in:

PGDNN/Code/iGEN-NET/network.py

Line 142 in f775d0e

# transform the x and y dim first

Is there any specific reason for that? Wouldn't it be possible to perform the 3d transform in one go?
Also I can't find the supplementary materials. It would be good if you can point to that.

Thanks!

Ground truth for scale and translation

I wonder how was the ground truth for the scale and translation generated?
Can you provide some details? Thanks.