lpips-j - GitLite

KO | EN

gitlite — search

#json #nlp #generative-ai #cpp #hacktoberfest #gpt #privacy #kotlin #java #php

lpips-j

(★ 16)

Minimal JAX/Flax port of `lpips` supporting `vgg16`, with pre-trained weights stored in the 🤗 Hugging Face hub.

파일 탐색기

ZIP 다운로드

- - __init__.py
  - lpips.py
- img8.jpg
- img8_edited.jpg
- img_test.py
- lpips_test.py
.gitignore
LICENSE
pyproject.toml
README.md
setup.cfg
setup.py

# CDN으로 사용하기

jsDelivr

jsDelivr는 공개 GitHub 리포지토리를 별도 설정 없이 CDN으로 즉시 서빙합니다. 버전과 파일을 고르면 웹페이지에 바로 붙일 수 있는 링크와 예시 코드가 만들어집니다.

버전

파일

링크

예시

명령어 용어집

이 문서에서 사용된 명령어를 모아봤습니다. 낯선 명령어가 있다면 펼쳐서 확인해보세요.

🔍

`pip install`

설명 보기 ▼

Install Python packages.

pip install {{package1 package2 ...}}

Install one or more packages:

pip install {{package1 package2 ...}} {{[-U|--upgrade]}}

Upgrade all specified packages to the latest version, installing any that are not already present:

pip install {{package}}=={{version}}

Install a specific version of a package:

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

This is a minimal JAX/Flax port of `lpips`, as implemented in:
* [richzhang/PerceptualSimilarity](https://github.com/richzhang/PerceptualSimilarity/blob/31bc1271ae6f13b7e281b9959ac24a5e8f2ed522/lpips/lpips.py)
* [Taming Transformers](https://github.com/CompVis/taming-transformers/blob/master/taming/modules/losses/lpips.py)

Only the essential features have been implemented. Our motivation is to support VQGAN training for [DALL•E Mini](https://github.com/borisdayma/dalle-mini).

It currently supports the `vgg16` backend, leveraging the implementation in [`flaxmodels`](https://github.com/matthias-wright/flaxmodels/blob/main/flaxmodels/vgg/vgg.py).

Pre-trained weights for the network and the linear layers are downloaded from the [🤗 Hugging Face hub](https://huggingface.co/pcuenq/lpips-jax).

## Installation

1. Install JAX for CUDA or TPU following the instructions at https://github.com/google/jax#installation.
2. Install this package:
   ```
   pip install lpips-j
   ```

## Use

Inputs must be in the range `[-1, 1]`, and _not_ already normalized with ImageNet stats. (They are internally converted to `[0, 1]` and then normalized by [the underlying flax model](https://github.com/matthias-wright/flaxmodels/tree/main/flaxmodels/vgg#1-important-note).

```Python
x = PILToTensor()(Image.open("img8.jpg")).unsqueeze(0)
y = PILToTensor()(Image.open("img8_edited.jpg")).unsqueeze(0)

x = 2 * (x / 255.) - 1
y = 2 * (y / 255.) - 1
x = jnp.array(x).transpose(0, 2, 3, 1)
y = jnp.array(y).transpose(0, 2, 3, 1)

lpips = LPIPS()
params = lpips.init(key, x, x)
loss = lpips.apply(params, x, y)
```