i2-multimedia-lab / ipdae Goto Github PK

Hello,
thanks for sharing your code! I have tried to use it with the recommended data but the encoder crashes with an exception.
Running as:

python ./compress.py 'ModelNet40_pc_01_8192p/airplane/test/*.ply' './test' './model/K256' --K 25
And getting:

Traceback (most recent call last):
  File "/home/ichlubna/Workspace/IPDAE/./compress.py", line 56, in <module>
    ae.load_state_dict(torch.load(NET_PATH))
  File "/usr/lib/python3.11/site-packages/torch/nn/modules/module.py", line 2153, in load_state_dict
    raise RuntimeError('Error(s) in loading state_dict for {}:\n\t{}'.format(
RuntimeError: Error(s) in loading state_dict for AE:
	size mismatch for inv_pool.4.weight: copying a param with shape torch.Size([16384, 1024]) from checkpoint, the shape in current model is torch.Size([1536, 1024]).
	size mismatch for inv_pool.4.bias: copying a param with shape torch.Size([16384]) from checkpoint, the shape in current model is torch.Size([1536]).

Would you please help me to solve this?
Thanks

In the article, it is shown that a learnable context model is adopted in your method. However, I only find a rate loss for the learnable context model in your method to upate the parameters. So how to ensure that the learnable context model estimate the pmf correctly? In my opinion, the rate loss only push the entropy of the pmf to be lower but not let the pmf estimation to be accurater. Is there anything wrong?