Search

​

Searching Through an Index

Search allows you to find portions of a video through natural language. Results are categorized into three types: visual embeddings, transcriptions, and optical character recognition (OCR).

To search:

# Search with natural language
resp = moonshine.search(index="your-index-name", query="red car with people shaking hands inside", num_args=10)

# resp = {'status': 'complete', 'output': {'embeddings': match[], 'transcript': match[], 'ocr': match[]}}

Where index is required and the index you want to search, query is required and your search query, and num_args is optional and number of results you want back.

Every match object is an individual result that matches the query. A list of match objects is returned for every category.

A match object has the following data:

​

Searching With an Image

You can search through an index with an image. To search with an image:

resp = moonshine.search(index="your-index-name", image="./image.png")

# resp = {'status': 'complete', 'output': {'embeddings': match[], 'transcript': match[], 'ocr': match[]}}

​

Searching an Index Subset (Subsearch)

You can narrow your search to a specific subset of videos within an index by using the subindex parameter. This allows you to search only through selected videos instead of the entire index.

# Search only within specific videos
resp = moonshine.search(
    index="your-index-name",
    query="red car with people shaking hands inside",
    subindex=["DWHEDUEIOJOWII.mp4", "AJDIEUJDKSLW.mp4"],
    num_args=10
)

The subindex parameter accepts a list of video_ids as strings. All video_ids must belong to the same parent index. This is useful when you want to focus your search on specific videos rather than searching through the entire index.