open_clip
StyleCLIP
open_clip | StyleCLIP | |
---|---|---|
30 | 23 | |
8,924 | 3,917 | |
5.3% | - | |
8.1 | 0.0 | |
3 days ago | about 1 year ago | |
Jupyter Notebook | HTML | |
GNU General Public License v3.0 or later | MIT License |
Stars - the number of stars that a project has on GitHub. Growth - month over month growth in stars.
Activity is a relative number indicating how actively a project is being developed. Recent commits have higher weight than older ones.
For example, an activity of 9.0 indicates that a project is amongst the top 10% of the most actively developed projects that we are tracking.
open_clip
-
Binarize Clip for Multimodal Applications
The part of CLIP[1] that you need to know to understand this is that it embeds text and images into the same space. ie: the word "dog" is close to images of dogs. Normally this space is a high dimensional real space. Think 512-dimensional or 512 floating point numbers. When you want to measure "closeness" between vectors in this space cosine similarity[2] is a natural choice.
Why would you want to quantize values? Well, instead of using a 32-bit float for each dimension, what if you could get away with 1-bit? You would save you 31x the space. Often you'll want to embed millions or billions of pieces of text or images, so the savings represent a huge speed & cost savings and if accuracy isn't impacted too much then it could be worth it.
If you naively clip the floats of an existing model, it severely impacts accuracy. However, if you train a model from scratch that produces binary outputs, then it appears to perform better.
There is one twist. Deep learning models rely on gradient descent to train and binary output doesn't produce useful gradients. We use cosine similarity on floating point vectors and hamming distance on bit vectors. Is there a function that behaves like hamming distance but is nicely differentiable? We can then use this function during training and then vanilla hamming distance during inference. It seems like they've done that.
I'd suggest playing around with OpenCLIP[3]. My background is in data science but all my CLIP knowledge comes from doing a side project over the course of a couple weekends.
1. https://huggingface.co/docs/transformers/model_doc/clip
2. https://en.wikipedia.org/wiki/Cosine_similarity
3. https://github.com/mlfoundations/open_clip
- FLaNK-AIM: 20 May 2024 Weekly
- FLaNK AI Weekly for 29 April 2024
-
A History of CLIP Model Training Data Advances
While OpenAI’s CLIP model has garnered a lot of attention, it is far from the only game in town—and far from the best! On the OpenCLIP leaderboard, for instance, the largest and most capable CLIP model from OpenAI ranks just 41st(!) in its average zero-shot accuracy across 38 datasets.
-
How to Build a Semantic Search Engine for Emojis
Whenever I’m working on semantic search applications that connect images and text, I start with a family of models known as contrastive language image pre-training (CLIP). These models are trained on image-text pairs to generate similar vector representations or embeddings for images and their captions, and dissimilar vectors when images are paired with other text strings. There are multiple CLIP-style models, including OpenCLIP and MetaCLIP, but for simplicity we’ll focus on the original CLIP model from OpenAI. No model is perfect, and at a fundamental level there is no right way to compare images and text, but CLIP certainly provides a good starting point.
-
Database of 16,000 Artists Used to Train Midjourney AI Goes Viral
It is a misconception that Adobe's models have not been trained on copyrighted work. Nobody should be repeating their marketing claims.
Adobe has not shown how they train the text encoders in Firefly, or what images were used for the text-based conditioning (i.e. "text to image") part of their image generation model. They are almost certainly using CLIP or T5, which are trained on LAION2b, an image dataset with the very problems they are trying to address, C4 (a text dataset similarly encumbered) and similar.
I welcome anyone who works at Adobe to simply answer this question of how they trained the text encoders for text conditioning and put it to rest. There is absolutely nothing sensitive about the issue, unless it exposes them in a lie.
So no chance. I think it's a big fat lie. They'd have to have made some other scientific breakthrough, which they didn't.
Using information from https://openai.com/research/clip and https://github.com/mlfoundations/open_clip, it's possible to investigate the likelihood that using just their stock image dataset, can they make a working text encoder?
It's certainly not impossible, but it's impracticable. On 248m images (roughly the size of Adobe Stock), CLIP gets 37% on ImageNet, and on the 2000m from LAION, it performs 71-80%. And even with 2000m images, CLIP is substantially worse performing than the approach that Imagen uses for "text comprehension," which relies on essentially many billions more images and text tokens.
-
MetaCLIP – Meta AI Research
https://github.com/mlfoundations/open_clip/blob/main/docs/op...
-
COMFYUI SDXL WORKFLOW INBOUND! Q&A NOW OPEN! (WIP EARLY ACCESS WORKFLOW INCLUDED!)
in the modal card it says: pretrained text encoders (OpenCLIP-ViT/G and CLIP-ViT/L).
-
Is Nicholas Renotte a good guide for a person who knows nothing about ML?
also, if you describe your task a bit more, we might be able to direct you to a fairly out-of-the-box solution, e.g. you might be able to use one of the pretrained models supported by https://github.com/mlfoundations/open_clip without any additional training
-
Generate Image from Vector Embedding
It says on the Stable Diffusion Github repo that it uses the “OpenCLIP-ViT/H” https://github.com/mlfoundations/open_clip model as a text encoder, and from my prior experience with CLIP, I have found that it is very easy to generate image and text embeddings (because CLIP is a multimodal model).
StyleCLIP
-
A History of CLIP Model Training Data Advances
While CLIP on its own is useful for applications such as zero-shot classification, semantic searches, and unsupervised data exploration, CLIP is also used as a building block in a vast array of multimodal applications, from Stable Diffusion and DALL-E to StyleCLIP and OWL-ViT. For most of these downstream applications, the initial CLIP model is regarded as a “pre-trained” starting point, and the entire model is fine-tuned for its new use case.
-
[D] What is the largest / most diverse GAN model currently out there?
I'm currently building a fork for StyleCLIP global directions which allows you to control multiple semantic parameters simoultaneously to generate and edit an image with StyleGAN and CLIP in realtime. I want to showcase its potential as a design tool. Unfortunately, GAN weights are trained on very domain-specific (faces, cars, churches) data. This makes them inferior to modern diffusion models which I can use to generate whatever comes to mind. Although I know we won't have a GAN-based DALL-E counterpart anytime soon, I still would love to use my system with weights that can output a wide variety of things.
-
test
(Added Feb. 15, 2021) StyleCLIP - Colaboratory by orpatashnik. Uses StyleGAN to generate images. GitHub. Twitter reference. Reddit post.
- I am David Bau, and I study the structure of the complex computations learned within deep neural networks.
-
Dragon Age Origins Companions as Photorealistic People.
I used StyleCLIP. I purchased some Google Colab time to use their GPUs. I'll probably do some more later this week.
- Turning BDO characters into blursed people with AI
-
I used AI to generate real life for honor character faces
Link for Styleclip
-
AI-generated 'real' faces of CGI characters - description in comments
So, I watched this Corridor Crew video on generating realistic faces from CG characters, and I wanted to try it out on the RDR2 models. The github link for the original work is here. If you guys are interested I can generate the faces of more characters from RDR2 and RDR1. I can even try some from RD Revolver.
-
AI Generated Art Scene Explodes as Hackers Create Groundbreaking New Tools - New AI tools CLIP+VQ-GAN can create impressive works of art based on just a few words of input.
Combining these methods with CLIP allows you to generate images based on text. This one uses a face generator. https://github.com/orpatashnik/StyleCLIP
- [D] How to save latent code edited from StyleClip.
What are some alternatives?
CLIP - CLIP (Contrastive Language-Image Pretraining), Predict the most relevant text snippet given an image
encoder4editing - Official implementation of "Designing an Encoder for StyleGAN Image Manipulation" (SIGGRAPH 2021) https://arxiv.org/abs/2102.02766
DALLE-pytorch - Implementation / replication of DALL-E, OpenAI's Text to Image Transformer, in Pytorch
compare_gan - Compare GAN code.
taming-transformers - Taming Transformers for High-Resolution Image Synthesis
NVAE - The Official PyTorch Implementation of "NVAE: A Deep Hierarchical Variational Autoencoder" (NeurIPS 2020 spotlight paper)
Dreambooth-Stable-Diffusion - Implementation of Dreambooth (https://arxiv.org/abs/2208.12242) with Stable Diffusion
stylegan2-pytorch - Simplest working implementation of Stylegan2, state of the art generative adversarial network, in Pytorch. Enabling everyone to experience disentanglement
bitsandbytes - Accessible large language models via k-bit quantization for PyTorch.
pixel2style2pixel - Official Implementation for "Encoding in Style: a StyleGAN Encoder for Image-to-Image Translation" (CVPR 2021) presenting the pixel2style2pixel (pSp) framework
clip-retrieval - Easily compute clip embeddings and build a clip retrieval system with them
alias-free-gan - Alias-Free GAN project website and code