vision

🛠️ OpenAI Vision API

Implements: OpenAI vision API.

GPT-4 with Vision ... GPT-4V or gpt-4-vision-preview in the API ... model ... take in images and answer questions about them ... not stateful ... (more)

Example use on the images ingested from the traffic cameras in downtown Vancouver, using vancouver-watching 🌈,

@openai vision \
    "list some of the things that you see in these images." \
    - Davie,Bute,.jpg \
    $(@mlflow tags search area=vancouver,ingest,published \
        --log 0 \
        --count 1 \
        --offset 0) \
    --max_count 10 \
    --verbose 1

The images show street scenes captured by surveillance cameras at night, and each bears a timestamp and the text "CITY OF VANCOUVER" indicating they're from Vancouver. In the first image, you see a yellow traffic light indicating caution and part of a crosswalk painted in rainbow colors, likely symbolizing support for the LGBTQ+ community. In the second image, there's a view of a city street with pedestrians crossing, some wearing reflective vests, and the storefronts illuminated with bright lights. The third image shows a different angle of a similar street scene with the same vibrant crosswalk and vehicles' headlights creating streaks of light due to the camera exposure. Finally, in the fourth image, there's a view of parked cars and a clearly visible police vehicle, suggesting the presence of law enforcement nearby. The wet pavement suggests it may have rained, and the reflections imply the street is well-lit at night.