Premade models

There are a few pre-made models for some common applications. Using a premade model is the easiest way to get started with Decthings, but if you need something more specific than what the default models can provide, you will need to create your own model. This document lists some premade models and how to use them.

Image classifier
MiDaS depth estimation

Image classifier

This model is an image classifier which uses the EfficientNet V2 (L) architecture to assign labels to images. The model is trained on the dataset ImageNet, which contains over 14 million images in 1000 classes. When you provide an input image to this model it will calculate one probability for each of the 1000 classes and output the top ten classes and their probabilities.

Border collie (67.8%)

Collie (12.8%)

Japanese spaniel (0.4%)

Papillon (0.4%)

Shetland (0.3%)

Tennis ball (0.2%)

Eskimo dog, husky (0.1%)

English springer (0.1%)

Appenzeller (0.1%)

Groenendael (0.1%)

You can try this model by going to the model page, navigating to the "Deploy" tab and then clicking "Perform evaluation". This will allow you to upload an image in your browser.

To use this model from your application or devices, install an API client, create an API key, and use the following code as an example.

Language

import * as fs from 'fs'
import { DecthingsClient, Data, DataElement } from '@decthings/api-client';

async function main() {
    // Create a Decthings API client
    let apiKey = fs.readFileSync('./auth.txt').toString()
    let client = new DecthingsClient({ apiKey })

    // Construct the input data
    let image = fs.readFileSync('./image.png')
    let inputData = new Data([DataElement.image('png', image)])

    try {
        // Evaluate the model
        let response = await client.model.evaluate({
            modelId: 'e3298f45-11db-488b-8aba-4ad513ac25f1',
            params: [{ name: 'input', data: inputData }]
        })
        if (response.error) {
            // Decthings did not accept the request for some reason.
            console.error('Could not start evaluation:', response.error)
        }
        else if (response.result.failed) {
            // The evaluation was started but failed, for example due to an exception in the model code.
            console.error('The evaluation failed:', response.result.failed)
        }
        else {
            let outputs = response.result.success.outputs
            // This model provides just a single output, so read it using [0]
            let outputData = outputs[0].data
            for (let classData of data.values()) {
                let className = classData.get('class').getString()
                let probability = classData.get('probability').getNumber()
                // Here we just log the output, but you can of course do something else with this information.
                console.log(`${className}: ${probability}`)
            }
        }
    } catch (e) {
        // Client throws an error on connection issues
        console.error('The request failed: ', e)
    }
}
main()

MiDaS depth estimation

This model takes an image as input and uses the Intel ISL MiDaS architecture to generate a depth map from that image.