BerkeleyAutomation / gqcnn

Python module for GQ-CNN training and deployment with ROS integration.

Home Page:https://berkeleyautomation.github.io/gqcnn

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Issue: Other Issue

futurev opened this issue · comments

commented

System information

  • OS Platform and Distribution (e.g., Linux Ubuntu 16.04): windows10
  • Python version: python3.6
  • Installed using pip or ROS: pip
  • GPU model (if applicable):

Describe the result you are trying to replicate
If you can, provide a link to the section in the documentation.

Provide the exact sequence of commands / steps that you executed to replicate this result

Describe the unexpected behavior

Other info / logs
Include any logs or source code that would be helpful to diagnose the problem. If including tracebacks, please include the full traceback. Large logs and files should be attached.

In the paper, Fig2 shows that fc3's kernel shapes is [18, 18], but when I run the code, I got filter_dim for fc3 [17, 17] in fc_network_tf.py--> _build_im_stream(), from input size 96*96--kernel (9, 9)--kernel (5, 5) --kernel (5, 5) --kernel (5, 5)
Could you please confirm which one is correct?

Hi @futurev,

What you are seeing at runtime is correct. There was an error in the paper calculation where we didn't use floor div, hence the off-by-one.

Thanks,
Vishal

commented

same thinking, thanks