Doug Morrison

Hi Olivia, I never labelled any of the data myself. I used either the existing Cornell dataset (which has one point of the rectangle per line ` `x y`) or...

It looks like the code assumes you're training on GPU. You can change https://github.com/dougsm/ggcnn/blob/0c50aa7600e8a30d44c5c85cebd6e3394a81f30e/train_ggcnn.py#L228 to "cpu" instead and it should work on your computer, but it will be quite slow...

Hi @trungpham2606 Sorry about that. I've uploaded new weights files (both the full pytorch model and the state dicts) and added instructions to the README. Please make sure you pull...

As you say, I think this is a symptom of a larger problem. This is only possible if there's not a single value stored anywhere in the histogram, as it...

Hey @aPonza I've just had a look but unfortunately I can't find the original joint values for those two named poses. As you say, I guess they were probably both...

@seivazi ROS has all of the functionality built in to run on separate networked machines. It's as simple as building the nodes on the machine on which they will be...

Hi @seivazi Sorry for the slow reply and glad you got it working. As far as I know, there should be no reason to use the older versions of the...

The only changes that I would have made to the panda_moveit_config would be to add the named poses to the srdf, and nothing else. Regarding the other issues, I have...

The distance from previous view was an extra loss for the next best view calculation which favoured viewpoints which were different from the current viewpoint (hence the x,y,z distance calculation),...

@seivazi @aPonza I have replied in the other issue which was about tensorflow versions. (I have also pushed some small updates that fix things that you were discussing). Are the...