New to Peltarion? Discover our deep learning Platform
A single deep learning platform to build and deploy your projects, even if you’re not an AI superstar.
FIND OUT MORESingle-label image classification / cheat sheet
Target audience: Data scientists and developers
Use this cheat sheet
If your input data consists of labeled images containing exactly one of multiple classes. This is called single-label classification.
Example use cases
|
Note
|
Disclaimer Please note that data sets, models and other content, including open source software, (collectively referred to as "Content") provided and/or suggested by Peltarion for use in the Platform, may be subject to separate third party terms of use or license terms. You are solely responsible for complying with the applicable terms. Peltarion makes no representations or warranties about Content and specifically disclaim all responsibility for any liability, loss, or risk, which is incurred as a consequence, directly or indirectly, of the use or application of any of the Content. |
|
Images of skin lesions that are either benign or malignant. |
|
Images of solar panels that are either defective or functional. |
|
Images that contain exactly one handwritten number from
0 to 9. |
Data preparation
You can add your dataset in several different ways in the Datasets view; Import from your data warehouse, directly from your local computer, via data API, or from an URL. Here we’ll show you how to add it directly from your computer.
Data input requirements
Prepare a zip file with a folder containing all your images and a corresponding index.csv.
The Peltarion Platform supports .jpg or .png.
Structure of index.csv
The index.csv file is a simple text file where columns are separated by a comma: ",". Each column will be imported as a dataset feature, where the name of the feature is taken from the first line.
You need to give at least two features: the images and their target classes. To specify the images, write the name of the image file, together with their paths inside the zip file if they are located in subfolders.
| image | label |
|---|---|
images/image_1.jpg |
class 1 |
images/image_2.jpg |
class 2 |
Modeling
Use the Experiment wizard to choose the best snippet. Snippets are pre-built neural network architectures available on the platform. Several are good for image classification.
| Image size | Recommended snippet |
|---|---|
1-31 pixels |
CNN |
32-223 pixels |
MobileNetV2 0.35 |
224 - 332 pixels |
EfficientNetB0 |
333 - 400 pixels |
EfficientNetB4 |
401 - 499 pixels |
EfficientNetB5 |
500 - 560 pixels |
EfficientNetB6 |
Try the smallest depth model first
Try the smallest depth model first, since it will be faster to train and may already be complex enough to model your data well.
If the results are not good enough, you can move towards increasingly deep models in later experiments.
Initialize weights
If available you should use pretrained weights where the snippet has already learned the basic representations from the dataset it has been trained on and you can now train it on a small domain-specific dataset to provide value.
Run experiment
When you’ve created an experiment with the Experiment wizard, a ready-to-run model will populate the Modeling canvas. Everything needed will be set, input and target feature, weights (if you wanted those), loss function, activation, runtime settings, i.e., batch size, learning rate, number of epochs.
You can still edit/add/remove anything on the Modeling canvas or override all the existing choices, delete the snippet, and add a new one.
The only thing left to do is to press the Run button on the upper right corner to start the training process.
Deployment view
In the Deployment view, click New deployment.
Select your experiment and epoch for deployment.
Click the Enable switch to deploy the experiment.
Image Classifier web app
To make it easier to test your model, we have provided a Image Classifier web app.
Open the Image classifier: bit.ly/ImageClassifier, and copy-paste the following information from the Deployment view:
-
Deployment URL
-
Token
-
input parameter feature name.
Then add an image and press the Run button to predict!
Example:
