Running Inference

Files, streams, URLs, and image groups

endpoint.process() accepts a source and resolves to a ResultStream, an AsyncIterable of predictions.

This page is the Node call shapes. What a source is, every kind the platform accepts, and the options that shape how one is processed are covered once in Sources and Options.

Local files

const results = await endpoint.process({ source: { path: 'image.jpg' } })
for await (const result of results) {
    console.log(result)
}

Readable streams

Streams need an explicit MIME type.

import fs from 'node:fs'
import { Readable } from 'node:stream'

const stream = Readable.toWeb(fs.createReadStream('image.jpg'))
const results = await endpoint.process({
    source: { stream, mimeType: 'image/jpeg' },
})

Public URLs

A URL is fetched by the platform, so nothing uploads from your application. Source Types lists every scheme it accepts.

const results = await endpoint.process({
    source: { url: 'https://example.com/video.mp4' },
})

Image groups

A group is one source processed together as a single inference unit, returning one prediction for the whole set.

// local files
const fileGroupResults = await endpoint.uploadGroup(['a.jpg', 'b.jpg', 'c.jpg'])

// readable streams, with optional parallel MIME types
import fs from 'node:fs'
import { Readable } from 'node:stream'

const a = Readable.toWeb(fs.createReadStream('a.jpg'))
const b = Readable.toWeb(fs.createReadStream('b.jpg'))
const streamGroupResults = await endpoint.uploadStreamGroup([a, b], ['image/jpeg', 'image/jpeg'])

// remote URLs
const urlGroupResults = await endpoint.loadFromGroup([
    'https://example.com/a.jpg',
    'https://example.com/b.jpg',
])

Image groups covers the size limit, the ordering guarantee, and which abilities accept a group.

Canceling jobs

Queued and in-progress jobs can be cancelled from the result iterator.

const results = await endpoint.process({
    source: { url: 'https://example.com/video.mp4' },
})

for await (const result of results) {
    console.log(result)
    if ((result.seconds ?? 0) >= 10) {
        results.cancel()
    }
}

Camera calibration

process() and the group methods take a camera, which is what lets a depth map become positions in meters:

const results = await endpoint.process({
    source: { url: 'rtsp://camera.example.com/stream1' },
    camera: { hfovDegrees: 72 },
})

Set it once for every source with pop.defaults instead — see Composable Pops.

Next steps

• Sources and Options — every source the platform accepts, and the options that shape processing

• Composable Pops — chain models into a pipeline

• Visualization — draw predictions on a canvas

• Depth and World Coordinates — depth maps, calibration, and meters