A containerized REST API around OpenAI's CLIP model.

Last update: Nov 06, 2022

Overview

OpenAI's CLIP — REST API

This is a container wrapping OpenAI's CLIP model in a RESTful interface.

Running the container locally

First, build the container:

docker build -t clip-container:latest .

Then, you can run it:

docker run -it -p 8080:8080 --name "clip-container" --rm clip-container:latest /opt/ml/code/serve

Sending requests:

The container exposes two different endpoints:

GET /ping: Returns 200 status if the container is working properly.
POST /invocations: Processes a list of images and returns the list of labels with their corresponding probabilities.

Here is an example request assuming the container is listening in port 8080:

curl --location --request POST 'http://localhost:8080/invocations' \
--header 'Content-Type: application/json' \
--data-raw '{
    "images": [
        "https://images.unsplash.com/photo-1597308680537-1ba44407ffc0?ixid=MXwxMjA3fDB8MHxwaG90by1wYWdlfHx8fGVufDB8fHw%3D&ixlib=rb-1.2.1&auto=format&fit=crop&w=1834&q=80",
        "https://images.unsplash.com/photo-1589270216117-7972b3082c7d?ixid=MXwxMjA3fDB8MHxwaG90by1wYWdlfHx8fGVufDB8fHw%3D&ixlib=rb-1.2.1&auto=format&fit=crop&w=1834&q=80"],
    "classes": ["person", "bag", "person with a bag", "woman riding a horse", "woman with a bag", "woman with black shirt and a bag"]
}'

The response looks like this:

[
    {
        "url": "https://images.unsplash.com/photo-1597308680537-1ba44407ffc0?ixid=MXwxMjA3fDB8MHxwaG90by1wYWdlfHx8fGVufDB8fHw%3D&ixlib=rb-1.2.1&auto=format&fit=crop&w=1834&q=80", 
        "labels": [
            "woman with black shirt and a bag", 
            "woman with a bag", 
            "person with a bag", 
            "bag", "person"
        ], 
        "probs": [1.0, 1.7488513970320696e-09, 1.1663764917350243e-19, 4.179975909038141e-30, 3.77612043676229e-30]
    }, 
    {
        "url": "https://images.unsplash.com/photo-1589270216117-7972b3082c7d?ixid=MXwxMjA3fDB8MHxwaG90by1wYWdlfHx8fGVufDB8fHw%3D&ixlib=rb-1.2.1&auto=format&fit=crop&w=1834&q=80", 
        "labels": [
            "person with a bag", 
            "woman with black shirt and a bag", 
            "bag", 
            "woman with a bag", 
            "person"
        ], 
        "probs": [1.0, 2.4879632576357835e-08, 2.065714813830402e-13, 7.658033346455602e-15, 1.1307645811408335e-23]
    }
]

SageMaker Integration

This container is compatible with SageMaker so you should be able to host it as a SageMaker endpoint with no modifications. The code supports GPU and CPU instances.

A containerized REST API around OpenAI's CLIP model.

Related tags

Overview

OpenAI's CLIP — REST API

Running the container locally

Sending requests:

SageMaker Integration

Owner

Santiago Valdarrama

Code for the paper "Unsupervised Contrastive Learning of Sound Event Representations", ICASSP 2021.

Code for "Intra-hour Photovoltaic Generation Forecasting based on Multi-source Data and Deep Learning Methods."

Rotation-Only Bundle Adjustment

Code for paper Novel View Synthesis via Depth-guided Skip Connections

Pixray is an image generation system

An implementation of the Contrast Predictive Coding (CPC) method to train audio features in an unsupervised fashion.

[WACV 2022] Contextual Gradient Scaling for Few-Shot Learning

Find-Lane-Line - Use openCV library and Python to detect the road-lane-line

Equivariant Imaging: Learning Beyond the Range Space

NeuralWOZ: Learning to Collect Task-Oriented Dialogue via Model-based Simulation (ACL-IJCNLP 2021)

Example scripts for the detection of lanes using the ultra fast lane detection model in Tensorflow Lite.

This is the official Pytorch implementation of "Lung Segmentation from Chest X-rays using Variational Data Imputation", Raghavendra Selvan et al. 2020

reimpliment of DFANet: Deep Feature Aggregation for Real-Time Semantic Segmentation

Official repository of the paper Privacy-friendly Synthetic Data for the Development of Face Morphing Attack Detectors

This demo showcase the use of onnxruntime-rs with a GPU on CUDA 11 to run Bert in a data pipeline with Rust.

This is a computer vision based implementation of the popular childhood game 'Hand Cricket/Odd or Even' in python

Tools for manipulating UVs in the Blender viewport.

Code for IntraQ, PyTorch implementation of our paper under review

A pytorch reproduction of { Co-occurrence Feature Learning from Skeleton Data for Action Recognition and Detection with Hierarchical Aggregation }.

Pop-Out Motion: 3D-Aware Image Deformation via Learning the Shape Laplacian (CVPR 2022)