clinplayer / 3dmodelingrl Goto Github PK

Have you test the performance on more complex categories? Such as chair, which involves hole and non-convex part.

Hi,

Thank you for sharing codes.

I wonder how you generate ground truth voxels.
I would like to run this code in another class of the ShapeNet dataset.
I have used binvox to generate a 64^3 voxel model from ShapeNet v2, but it does not match the model you are providing.
Specifically, the following commands were executed.
binvox model_normalized.obj -d 64 -cb

If there is no problem with the command, it is because of the version of binvox or my environment. So I won't worry about this anymore.
If you know of a better way, please let me know.

hi,
I try to download the dataset and pretrained models, but the google drive does not allow me to view them and show to be "access denied".
could you please update the download link?
thanks!

Hi Lin,

thanks for sharing your interesting and impressive work!
I am trying to implement it on my own machine. Could you please tell me how long it took you to train both agents and hardware specifications of the machine you used?

Thanks in advance!

Best,
yudopan

Hi,
thx for sharing the source code of this wonderful project.
I have a question following issue #1 .
I am recently training both Prim and Mesh agents on airplane dataset using the depth image input.
However, it took very long time and I am not sure whether the training is really effective.
So far, for Prim agent, each epoch took 0.85 hours, and I am currently running until 140 epochs (more than 4 days as you mentioned in #1 )
However, the reward curves are not converging based on my observation.
And I compared the currently trained model with the pretrained Prim agent weights (compare the mean rewards using test.py).
The mean reward of the currently trained model is still pretty bad.
I am wondering do you happen to have any ideas on potential mistakes I made here?

thank you