tensorflow / tensorflow/probability
Feature Request: Vector Quantizer Layer
Nobody has claimed this yet.
- Dominant language
- Jupyter Notebook
- Stars
- 4.4k
- Forks
- 1.1k
- PR merge metrics
- No merged PRs in 30d
Description
A recent paper(1) achieving state of the art results on image generation, has sparked a lot of interest in VQ research, but for quite complex architectures (this paper uses a three-stage VQ), making a "layer" abstraction of a VectorQuantizer quite a valuable tool.
Properties of implementation:
Ideally it would work for any dtype/rank tensor.
It should have a call method that is used for training. (or?)
It should have a method to decode discrete tokens, accepting a int32 tensor, looking up codebook
It should have a method to encode continuous latents, that performs the nearest neighbours calculation on the codebook, returning a int32 tensor.
Any more ideas?
- Generating Diverse High-Fidelity Images with VQ-VAE-2
https://arxiv.org/abs/1906.00446v1
Contributor guide
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
Research direction
Start by reviewing the requested VectorQuantizer behavior and the cited VQ-VAE-2 paper. Clarify the layer API and tensor-shape requirements before locating an appropriate TensorFlow Probability entry point. Done means supporting training, encoding continuous latents to int32 tokens, and decoding tokens through a codebook across the requested tensor dtypes and ranks.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- tensorflow
- Domain
- machine-learning
- Issue type
- Feature
- Difficulty
- 5/5
- Estimated time
- Over a week
- Activity status
- Stale
- Clarity
- Needs clarification
- Newbie friendliness
- 25/100