OpenAI CLIP

CLIP (Contrastive Language-Image Pre-Training) is an impressive multimodal zero-shot image classifier that achieves impressive results in a wide range of domains with no fine-tuning. It applies the recent advancements in large-scale transformers like GPT-3 to the vision arena.

Try the Model

Use the widget below to experiment with OpenAI CLIP. You can detect COCO classes such as people, vehicles, animals, household items.

Overview

What is CLIP?

In January 2021 OpenAI released CLIP (Contrastive Language-Image Pre-Training), a zero-shot classifier that leverages knowledge of the English language to classify images without having to be trained on any specific dataset. It applies the recent advancements in large-scale transformers like GPT-3 to the vision arena.

The results are extremely impressive; we have put together a CLIP tutorial and a CLIP Colab notebook for you to experiment with the model on your own images.

CLIP's Performance

Training Efficiency: CLIP is among one of the most efficient models with an accuracy of 41% at 400 million images, outperforming other models such as the Bag of Words Prediction (27%) and the Transformer Language Model (16%) at the same number of images. This means that CLIP trains much faster than other models within the same domain.

Generalization: CLIP has been trained with such a wide array of image styles that it is far more flexible and than other models like ImageNet. It is important to note that CLIP generalizes well with images that it was trained on, not images outside of its training domain. Pictured below are some of the different image styles:

OpenAI CLIP License

OpenAI CLIP

is licensed under a

MIT

license.

Performance

Deploy a OpenAI CLIP API

You can use Roboflow Inference to deploy a

OpenAI CLIP

API on your hardware. You can deploy the model on CPU (i.e. Raspberry Pi, AI PCs) and GPU devices (i.e. NVIDIA Jetson, NVIDIA T4).

Below are instructions on how to deploy your own model API.

First, install Inference:


pip install inference

To classify images with CLIP, use the following code:


import os
from inference_sdk import InferenceHTTPClient

CLIENT = InferenceHTTPClient(
    api_url="https://infer.roboflow.com",
    api_key=os.environ["ROBOFLOW_API_KEY"],
)

result = CLIENT.clip_compare(
  subject="./image.jpg",
  prompt=["dog", "cat"]
)
print(result)

Label Data Automatically with OpenAI CLIP

You can automatically label a dataset using OpenAI CLIP with help from Autodistill, an open source package for training computer vision models. You can label a folder of images automatically with only a few lines of code. Below, see our tutorials that demonstrate how to use OpenAI CLIP to train a computer vision model.

CLIP

CLIP and YOLOv5

CLIP and YOLOv8