blip-image-captioning-docker's Introduction

BLIP Image Captioning inside a Docker Container with simple API

Features

Worker Queue
Create Image Caption of Image by passing URL

API Endpoints

/captionImage GET
- Get job results
/captionImage POST
- Post a job to the queue
- Takes json with imgurls: ['url1', 'url2', ...]
- When job is done the result will be added to the result list
/captionImageBlocking POST
- Post a job to the queue
- Takes json with "imgurls": ['url1', 'url2', ...]
- Waits for job to finish and returns the result in the response

Setup/Installation

I assume that you have docker installed and a CUDA capable GPU
I suggest that you run everything locally first to verify that every thing works as the docker image build can take quite long
After running it locally for the first time, there should be a /checkpoints folder with the BLIP model
- So the docker container can copy it and doesnt need to download it again

Example

POST /captionImage

Request:

{
    "imgurl": ["https://images.pexels.com/photos/45201/kitty-cat-kitten-pet-45201.jpeg?auto=compress&cs=tinysrgb&w=1600"]
}

Response:

{
    "id": "4fbc68319cd6491daf7451139f58b6c7",
    "msg": "Job Started!"
}

POST /captionImageBlocking

Request:

{
    "imgurl": ["https://images.pexels.com/photos/45201/kitty-cat-kitten-pet-45201.jpeg?auto=compress&cs=tinysrgb&w=1600", "https://images.pexels.com/photos/45201/kitty-cat-kitten-pet-45201.jpeg?auto=compress&cs=tinysrgb&w=1600"]
}

Response:

{
    "id": "93496917b71b44d49138cab7e4c5451d",
    "msg": "Job finished!",
    "result": [
        "a black and white cat sitting in the snow",
        "a black and white cat sitting in the snow",
        "a black and white cat sitting in the snow",
        "a black and white cat sitting in the snow",
        "a black and white cat sitting in the snow",
        "a black and white cat sitting in the snow",
        "a black and white cat sitting in the snow",
        "a black and white cat sitting in the snow",
        "a black and white cat sitting in the snow",
        "a black and white cat sitting in the snow"
    ]
}

GET /captionImage

Response:

[
    {
        "id": "4fbc68319cd6491daf7451139f58b6c7",
        "result": [
            "a white kitten with blue eyes sitting on a leopard print blanket"
        ]
    }
]

blip-image-captioning-docker's People

Contributors

Stargazers

Watchers

blip-image-captioning-docker's Issues

checkpoints missing

When running the powershell file
docker build -t py-worker .
it fails on
#7 [2/7] ADD ./checkpoints /app/checkpoints
#7 sha256:66385eb2b9e14ebca5144f4dc9ed2f54ac7cfa4adfaee182aaf2035af0103d3b
#7 ERROR: "/checkpoints" not found: not found

the readme says:

After running it locally for the first time, there should be a /checkpoints folder with the BLIP model
- So the docker container can copy it and doesnt need to download it again

It seems that there is step missing somewhere.

Shall I just drop model_base_caption_capfilt_large.pth in the folder?

Recommend Projects