aluo-x / 3d_sln Goto Github PK

Hello, thank you very much. Your work has been of great help to me. I used my datasets to train the network model and got KL_Loss = 0.05. However, when I use the same scene graph to guide the synthesis of scene layouts, I can only get the same generated scene graphs (little difference). why? Did I miss an important step?

for k in range(Nsample):
z = torch.from_numpy(np.random.multivariate_normal(mean_est, cov_est, objs.size(0))).float().cuda()
boxes_pred, angles_pred = model.decoder(z, objs, triples, attributes)

Nsample = 4 , Four boxes_preds are the same(little difference).

hi awesome work, can you please share more details about the section 5.2. Sentence-Based Scene Layout Synthesis in the paper? how do you parse the sentence and transform into scene graph (maybe ?) and generate the layout? thanks.

Hello, your article is really wonderful, I am also very interested in your work, I have tried to generate a reasonable indoor 3D layout using Sun RGB-D data. But I don't know the specific implementation project, can you give some guidance?

https://u.pcloud.link/publink/show?code=XZaDaNkZ4bapJodBqTjUBrpyMqUT8zqqCQHk