Image Search#
This application compares the image embedding calculated from the images in the rosbag with the text embedding calculated from the prompts entered by the user using various Vision&Language models, and presents the closest result as the search result.
In order to use this application, please setup amber with
poetry install --with apps
This application uses qdrant, a type of vector search engine, for searching.
1- shows the processing steps up to the point where the embedding of the image is calculated and it is registered in qdrant. 2- indicates the point where prompt input is accepted and displayed.
Run application with ford dataset.#
Warning
This sample command is written with the assumption that it will be executed in the root directory of the amber package.
python3 amber/apps/image_search.py --rosbag_directory tests/rosbag/ford/ --sampling_duration=0.1
If it works correctly, the following message is displayed.
Running on local URL: http://127.0.0.1:7860
To create a public link, set `share=True` in `launch()`.
Please access this URL as soon as the message is confirmed.
You can search for images by entering a prompt in gradio's UI as shown in the video below.