Top-quality data
Collect and annotate training data that meets and exceeds industry quality standards thanks to multiple quality control methods and mechanisms available in Toloka.
Scalable projects
Have any amounts of image, text, speech, audio or video data collected and labeled for you by millions of skilled Toloka users across the globe.
Cost-efficiency
Save time and money with this purpose-built platform for handling large-scale data collection and annotation projects, on demand 24/7, at your own price and within your timeframe.
Free, powerful API
Build scalable and fully automated human-in-the-loop machine learning pipelines with a powerful open API and Python library for easy integration.
Annotations we support
With Toloka, you can control data labeling accuracy to build a predictable pipeline of high-quality training data that impacts your CV algorithms. Our platform supports annotation for image classification, semantic segmentation, object detection and recognition, and instance segmentation. Labeling tools include bounding boxes, polygons and keypoint annotation.
1
Segmentation
2
Bounding box
3
Polygon
4
Keypoint
5
Сlassification
6
Image transcription
7
Side-by-side comparison
8
Image & video collection
Ask Tolokers to draw pixelwise boundaries for each object instance or group of objects.
Price for 1000 tasks: $15. Turnaround time: 3 hours.*
Ask Tolokers to identify objects in images that match certain classes and use bounding boxes to mark the location.
Price for 1000 tasks: $15. Turnaround time: 3 hours.*
Ask Tolokers to identify objects in images that match certain classes and draw pixel-perfect polygons around the exact shape.
Price for 1000 tasks: $15. Turnaround time: 3 hours.*
Tolokers can label feature details in human faces to identify facial landmarks, expressions or emotions.
Price for 1000 tasks: $15. Turnaround time: 3 hours.*
Use Toloka to classify images by matching visual content with one or more pre-defined categories.
Price for 1000 tasks: $4.5. Turnaround time: 1 hour.*
Ask Tolokers to transcribe text in PDF files. Use labeled data to train your text recognition algorithms, or validate and fine-tune the output of your own OCR models.
Price for 1000 tasks: $15. Turnaround time: 3 hours.*
Use side-by-side image comparisons to verify or clean up your data. Ask Tolokers to look at two images and pick which one is better.
Price for 100 image pairs starts at $1.5. Turnaround time: 1 hour.*
Collect datasets of videos or images related to a common theme, or with a specific type of lighting or environment. Ask Tolokers to record video snippets according to your specifications and instructions.
Price for 100 tasks: $6. Turnaround time: 5 hours.*
* Approximate cost. Not a public offer. Price and turnaround time for tasks are set by the requester and depend on the type of task, input data, and other factors.
Real-time insights
Track your projects with real-time statistics on progress, spending, quality, time spent on tasks and active users involved. Leverage detailed analytics to fine-tune as necessary and make timely decisions to optimize speed, quality and budget.
Success stories
Alexey Hahunov, founder and CTO of Dbrain, IT company and Y Combinator alumnus:
“Toloka helped us tackle the challenge of recognizing handwritten text in documents. We send anonymous text fragments to Toloka, where qualified performers rewrite the text and send the results back to the Dbrain system for further AI training. This meant we were able to resolve even the most difficult recognition cases for our customers.”
Konstantin Simonchik, science director and co-founder of ID R&D:
“Using Toloka, we collected the world’s largest database of 200,000 unique photos and videos to secure biometric systems against hackers and train neural networks to distinguish real faces from fakes. We chose Toloka because of the user-friendly interface, active performers and optimal quality for the price.”
Dmitry Akimov, VisionLabs data engineer:
“Toloka is an excellent tool for data processing. We use it to collect and annotate thousands of images and videos every month in order to improve the performance of our algorithms.
We chose Toloka because of the fast turnaround time and the active participation of performers.”
Mikhail Burtsev, head of the DeepPavlov.ai laboratory for neural networks and deep learning at Moscow Institute of Physics and Technology:
“Modern machine learning technologies are not possible without a large volume of training examples. A computer can’t master a natural language unless a person ‘explains’ how to use it correctly. Toloka allows the conversational AI systems that we develop at MIPT to communicate with a large group of people and grow smarter.”
Angelina Galkevich, Analyst Tinkoff:
“We use Toloka to annotate data for speech synthesis and recognition Toloka performers have already processed tens of thousands of tasks with audio files for us, and we have successfully used the results to train our models. We like the price, the fast turnaround, and the super fast feedback. It allows us to quickly test different markup methods and hypotheses, and then choose the optimal approach.”
Anton Slesarev, head of Technologies in the self-driving cars division at Yandex:
“Toloka is the first place we go to prepare data for AI. To train the neural network, we need tens of thousands of annotated images. You can buy them for $4 each, or you can have images marked up in Toloka 10 times cheaper and get datasets that are ready to use for training algorithms. We use these datasets for developing self-driving vehicle technology.”
Yandex.Maps Team:
“Our service tracks outdated information on Yandex.Maps. We update business hours, add new businesses, and remove any that have closed. Toloka users perform our tasks in their neighborhoods, like taking photos of business signs with operating hours.
Every month we distribute 15 million tasks like this to 50,000 performers — this is the only way we can keep the huge Yandex.Maps database up to date.”
Alexey Hahunov, founder and CTO of Dbrain, IT company and Y Combinator alumnus:
“Toloka helped us tackle the challenge of recognizing handwritten text in documents. We send anonymous text fragments to Toloka, where qualified performers rewrite the text and send the results back to the Dbrain system for further AI training. This meant we were able to resolve even the most difficult recognition cases for our customers.”
Konstantin Simonchik, science director and co-founder of ID R&D:
“Using Toloka, we collected the world’s largest database of 200,000 unique photos and videos to secure biometric systems against hackers and train neural networks to distinguish real faces from fakes. We chose Toloka because of the user-friendly interface, active performers and optimal quality for the price.”
Dmitry Akimov, VisionLabs data engineer:
“Toloka is an excellent tool for data processing. We use it to collect and annotate thousands of images and videos every month in order to improve the performance of our algorithms.
We chose Toloka because of the fast turnaround time and the active participation of performers.”
Mikhail Burtsev, head of the DeepPavlov.ai laboratory for neural networks and deep learning at Moscow Institute of Physics and Technology:
“Modern machine learning technologies are not possible without a large volume of training examples. A computer can’t master a natural language unless a person ‘explains’ how to use it correctly. Toloka allows the conversational AI systems that we develop at MIPT to communicate with a large group of people and grow smarter.”
Angelina Galkevich, Analyst Tinkoff:
“We use Toloka to annotate data for speech synthesis and recognition Toloka performers have already processed tens of thousands of tasks with audio files for us, and we have successfully used the results to train our models. We like the price, the fast turnaround, and the super fast feedback. It allows us to quickly test different markup methods and hypotheses, and then choose the optimal approach.”
Anton Slesarev, head of Technologies in the self-driving cars division at Yandex:
“Toloka is the first place we go to prepare data for AI. To train the neural network, we need tens of thousands of annotated images. You can buy them for $4 each, or you can have images marked up in Toloka 10 times cheaper and get datasets that are ready to use for training algorithms. We use these datasets for developing self-driving vehicle technology.”
Yandex.Maps Team:
“Our service tracks outdated information on Yandex.Maps. We update business hours, add new businesses, and remove any that have closed. Toloka users perform our tasks in their neighborhoods, like taking photos of business signs with operating hours.
Every month we distribute 15 million tasks like this to 50,000 performers — this is the only way we can keep the huge Yandex.Maps database up to date.”