search

dougsm / ggcnn

Generative Grasping CNN from "Closing the Loop for Robotic Grasping: A Real-time, Generative Grasp Synthesis Approach" (RSS 2018)
BSD 3-Clause "New" or "Revised" License
499 stars 140 forks source link

Is there a bug in models/ggcnn.py? #42

Closed OranHe closed 3 years ago

OranHe commented 3 years ago

Hi, I think the padding here is 4 instead of 3. https://github.com/dougsm/ggcnn/blob/0c50aa7600e8a30d44c5c85cebd6e3394a81f30e/models/ggcnn.py#L16

OranHe commented 3 years ago

In the paper, the first layer input size is 300x300x1c, kernel size is 9x9, and stride is 3. I assume the output size is 100x100x1c. So the padding size should be 4=(9-1)/2.

When padding size is 3, the output size will be 99x99x1c, it looks a bit strange. Can you explain why u use padding3 instead of padding4 to me? Thank u so much.

OranHe commented 3 years ago

Thank you, I've figured it out