You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
This implementation was transcribed from the official Tensorflow version <ahref="https://github.com/hojonathanho/diffusion">here</a>
This might be too strong of a claim, as there are too many architectural differences between the two implementations (differences in where attention is placed, different number of residual streams, different number of ResNet blocks in down vs up stages, missing dropout, different activation functions in places, to name a few). It might also be somewhat misleading, as one might expect to use this repository to reproduce the results from the original paper, but the architectures are substantially different (see #114, #192).
I would suggest to rephrase the README.md as to better convey that this is a working reimplementation with differences to the original codebase. For example, "inspired by the official implementation", might be a good way to convey this. Alternatively, the codebase could be aligned more closely with the original implementation.
The text was updated successfully, but these errors were encountered:
BrunoKM
changed the title
"transcribed from official implementation" -> "Inspired by official implementation"
"transcribed from official implementation" -> "Inspired by official implementation" in README.md
Feb 21, 2024
The
README.md
currently says that the codebase is transcribed from the original codebase:denoising-diffusion-pytorch/README.md
Line 7 in c59ebf4
This might be too strong of a claim, as there are too many architectural differences between the two implementations (differences in where attention is placed, different number of residual streams, different number of ResNet blocks in down vs up stages, missing dropout, different activation functions in places, to name a few). It might also be somewhat misleading, as one might expect to use this repository to reproduce the results from the original paper, but the architectures are substantially different (see #114, #192).
I would suggest to rephrase the
README.md
as to better convey that this is a working reimplementation with differences to the original codebase. For example, "inspired by the official implementation", might be a good way to convey this. Alternatively, the codebase could be aligned more closely with the original implementation.The text was updated successfully, but these errors were encountered: