search

nv-tlabs / ATISS

Code for "ATISS: Autoregressive Transformers for Indoor Scene Synthesis", NeurIPS 2021
Other
258 stars 55 forks source link

The special q hat token with dimension 512 #15

Closed JackW987 closed 2 years ago

JackW987 commented 2 years ago

Thanks for your wonderful work. In the paper,the special q hat token with the dimension 64 and concated to predict other info.But in the code ,the special q hat token with wrong dimension 512.

paschalidoud commented 2 years ago

HI @JackW987,

Thank you for your feedback! You are correct, the dimensionality of the $\hat{\mathbf{q}}$ should have been 512 also in the paper. Please note that I have tested $\hat{\mathbf{q}}$ on the bedroom scenes with various dimensions and I didn't note any significant difference in performance, hence even if you use 64 dimensions it should work fine!

However, I will try to update the paper to avoid confusion.

Best, Despoina