How to create a MLImageClassifier in Xcode 10

Question

I'm trying to build a Machine Learning - Image Recognition using Create ML in Xcode 10.1 Playground but I'm having some problems to put my data in the model.

I have a folder with images numbered from 1 to 1336 and a .csv file with 2 columns (the image name and the image classification).

I don't know exactly how to put this in the model.

I have this until now:

import Cocoa
import CreateML

let data = try MLDataTable(contentsOf: URL(fileURLWithPath: "/Users/x/Desktop/CoreML/project/file.csv"))

let(trainingData, testingData) = data.randomSplit(by: 0.8, seed: 1)

let Classifier = try MLImageClassifier *need help here*

let evaluationMetrics = sentimentClassifier.evaluation(on: testingData)

let evaluationAccuracy = (1 - evaluationMetrics.classificationError) * 100

let metaData = MLModelMetadata(author: "x", shortDescription: "Model", version: "1.0")

try classifier.write(to: URL(fileURLWithPath: "/Users/x/Desktop/CoreML/project/XClassifier.mlmodel"))

score 0 · Answer 1 · answered Feb 23 '19 at 16:21

0

I believe it is not possible to feed labels to MLImageClassifier via .csv or any other separate file. You have only two options: use file names as labels or use directories as labels (probably preferable in your case of many images):

let model = try MLImageClassifier(trainingData: .labeledDirectories(at: trainingDir))
let evaluation = model.evaluation(on: .labeledDirectories(at: testingDir))

You will need to put images into subdirectories named as labels in your .csv file.

answered Feb 23 '19 at 16:21

Maxim Volgin

3,957
1
23
38

Do you know if there is a way to do this automatically? A script or something else? – Leonardo Feb 23 '19 at 23:00
Sure you can write a script in Swift or any language (Python, bash, etc.) – Maxim Volgin Feb 24 '19 at 02:50

score 0 · Answer 2 · answered Jan 09 '23 at 14:28

I was just struggling with this myself. Here is a solution to re-organise data for CreateML. All credit goes to Tony T1 who came up with this script.

Place images and CSV file into a single folder.
In Automator, create a new workflow like this:

Run the workflow. Select your CSV and watch the images get sorted into their respective folders!

The script is as follows:

cd "${1%/*}"
while read line         
do         
     FolderName=${line%;*}
     ImageName=${line#*;}
     mkdir "$FolderName"
     mv "$ImageName" "$FolderName"
done < "$1"

DF20 is the starting folder, you can change that to whatever you wish
my CSV was separated by ";". If your CSV is separated by ",", change that symbol in the script (e.g. FolderName=${line%,*} )
In my CSV, classes were columnA and images columnB. Switch this around depending on your case.

How to create a MLImageClassifier in Xcode 10

2 Answers2