Custom Image Recognition






I want to:
Image Categorization & Tagging

MY OWN AI
Train custom Visual AI for image recognition
With the Categorization & Tagging service, anyone can define custom categories & tags, link them to uploaded training images, and train custom image recognition models.
This way, you can automate the quality control of your products, tagging, sorting, filtering, and even recommendations of the items or images from your collection.

DELEGATE ROUTINE TASKS TO AI
AI processes visual data consistently & efficiently in milliseconds
Big databases of pictures can be managed by AI running on Ximilar cloud, which can handle large volumes of data 24/7. You can access it via API and integrate it into your application.
Working with the Ximilar platform doesn’t require a specific skill set. You easily train, chain, and deploy your models or contact us, and we will take care of the setup.
Enrich
your data with detailed information on images
Delegate
routine tasks to consistent AI trained on your data
Save
time and resources with AI automation
WHAT IS CATEGORIZATION?
Assign a category to each image
Image categorization assigns each image a category, such as a maxi dress or midi dress. The categories are visually distinctive, and each image belongs only to one category.
WHAT IS TAGGING?
Tag every image with many tags
Define a set of tags for the features & objects that should be recognized in your images, and train a custom tagging model able to provide tags for each image in your collection.
Interested in ready-to-use solutions?
Did you know we also have ready-to-use recognition services, that can be used straight away?
Check out the services tailored for Fashion, Home decor, and Stock Photos.
Custom Object Detection

TRAIN AI TO SPOT ANY OBJECT
Object Detection finds & marks objects from different categories
At Ximilar’s platform, you can train custom object detection models to identify any object, such as people, cars, particles in the water, imperfections of materials, objects of the same shape, size, or color – or we can train them for you.
An object detection task can either work independently or in a combination with categorization & tagging tasks to tag the detected objects.
Q & A
How do I prepare the training data?
The training of object detection models requires bigger datasets and more training time. It begins with data annotation – the manual marking of objects with bounding boxes. You can use the same dataset as for Categorization & Tagging model training.
Q & A
How do you work with my data?
During the training of custom image recognition models, your annotated images are divided into two groups. Apart from the training set, there is a smaller validation set, which is used to evaluate the accuracy of the model before the deployment. You can also upload another independent test set.
One Platform – Two Interfaces
App
You can annotate your images in Ximilar App. Just click on the image and draw a bounding box.
TRY FOR FREEAnnotate
Annotate was built for the annotation of large image datasets with complex hierarchical taxonomies in a team.
READ MOREFlows: Chain & Combine the Models

A HIERARCHY OF MODELS
Divide a complex problem into separate recognition, detection & tagging tasks
With Flows, the machine learning models can be combined and chained in a sequence, that exactly matches your decision-making process while working with visual data.
A Flow is like a tree, that grows bigger with the hierarchy of your models. Each image travels through the tree until it is properly processed and tagged. Based on this hierarchy, you also get suggestions when annotating the images.

ENDLESS POSSIBILITIES
Make a network of models and change any part you need anytime
- Re-train, add or remove any unit anytime
- Combine custom and ready-to-use services
- Recognize exclusively the detected objects
- Call multiple tasks (models) in one API call
- Call multiple recognition tasks in parallel
- Call endless nested flows by a primary flow
- Use one flow in several places
Build rich hierarchy
Define a flow with a few clicks, then use it for both training & automation
Play with the features
Add, remove, or change components, duplicate & modify your flows
Make changes on the fly
Flow structure handles any changes to both dataset and connected models
EXAMPLE: REAL ESTATE
Conditional image processing
Imagine you are building a real estate website. The first models in your flow can filter out all images that don’t meet certain selection criteria. In this case, it would be the pictures without any real estate, rooms, or furnishing.
EXAMPLE: REAL ESTATE
Automatic filtering, sorting & tagging
Images can then be gradually sorted with an increasing level of precision. The first task separates apartments and houses. Then, the apartments are sorted by room type, design, and furniture decor, and the houses by features such as architecture, area, garden or swimming pool.
Image Regression

AI MODEL THAT PREDICTS VALUES
Predict size of objects, age of people, or rating from images
The image regression model is able to predict numerical values from a range that you define in your images.
Image regression is used for building quality control systems, and to estimate values such as age, size, worn out level, or rating. It is a great tool for real estate and insurance companies, MedTech, and Industry 4.0. Contact us to discuss your application!
API DOCUMENTATIONBe Ahead of the Competition
Unlimited number of images
There are no limits on number of images per model/label
Use one image for many models
You can use the same images for the training of different models
Built-in data augmentation
You don’t have to prepare or multiply the training data in advance
No paying for training time
Unlike the competition, Ximilar doesn’t charge you for the training time
No paying for idle time
No fees for the idle time either
Cashing deployed models
Image processing takes 300 ms, as opposed to 2-3 s at other platforms
TECHNOLOGY STACK
We use state of the art neural network models & machine learning techniques
Our AI is improving constantly, so you always have up-to-date technology. Each model has millions of parameters that can be processed by CPU or GPU.
Our intelligent algorithm picks and uses the best performing models. We are using the latest technologies for machine learning as TensorFlow or OpenVINO.
Tips & Tricks
API Documentation
We take care of the complexity behind and wrap it in a few lines of code.
FULL DOCUMENTATION