Future
starts
CVAL helps ML team to select data efficiently to improve computer vision models
here
HOW IT WORKS
We supply pipelines for iterative training of models using data that will make the greatest contribution to the quality of the model
Minimum effort, maximum quality of your model through active learning
ACTIVE LEARNING PIPLINE
Signals for updating the model
/05
Model version ready for deployment
/04
Model training
/03
Data labeling
/02
Image sampling using CVAL
/01
Pool of unlabeled images
our Advantages
CVAL makes it possible to get a model of better quality that requires less data due to using efficient strategies for sampling images from a pool of unlabeled ones
-50%
Reduce operating costs for labeling
Meet deadlines and budget for data labeling by reducing the amount of labeling
+2-3 p.p.
To the quality of a model by predictive budget
Gaining competitive advantages by overcoming the industry threshold for ML product quality
98%
Focusing highly qualified ML personnel on creating key product value
Increasing the efficiency of using the company's ML resources by focusing them on the core functionality of the product
Privacy First is a balance of flexibility and security. The model and data will remain on your side
Product features
Minimizing time to market and labeling costs while swiftly delivering and maintaining high-quality models
Customer can choose one of these options
CVAL's in-house computation allows for rapid hypothesis testing, streamlining your model iteration process
Comprehensive set of sampling strategies - will allow you to get the most useful data for training models
Client library for quickly implementing your own active learning pipelines
Including a pipeline to keep the model up to date by correcting data drift
Setting up an AL-pipeline in a few minutes
*For the Privacy First plan
Selection
Using a combination of model predictions and embeddings, start the process of selecting data for labeling
API-key
Get the API-key and install the client library
Settings
Configure sampling strategies
Training
Train a model based on selected data
01
02
03
04
Improvement
Carry out steps
3 and 4 until the required model quality is achieved
3 and 4 until the required model quality is achieved
05
OUR CASES
our clients
Largest metallurgical plant*
Analysis of opportunities for process optimization in the leading metallurgical company through the implementation of information technologies
*According to the terms of the NDA, we are not allowed to disclose
Largest railway carrier*
Approach to improving the efficiency of information technology systems within the largest railway carrier
*According to the terms of the NDA, we are not allowed to disclose
about us
Konstantin Suhorukov
CEO
We are a young startup with a clear business vision and energetic engineers
01
We have a dozen projects in the field of computer vision
02
Victories in major IT competitions in AI
03
Product management in ML
Experience in running his own business
More than 15 years in Tech
Kirill Belyakov
Backend Lead
Development of high-load Backend solutions in Python / Golang
Prize places in programming competitions
Important Employee
CTO
TechLead ML in CV
Prize-winning places in ML competitions
More than 12 years in Tech
pricing
/cloud
A data set of up to 100K frames
889
$/month
For an additional 100K frames
The dataset is loaded into the service, the target model (from predefined ones) is placed in the service. Embeddings are computed using a predefined embedder or through a custom embedder on the client's side
889
$/month
/Privacy first
A data set of up to 100K frames
445
$/month
For an additional 100K frames
Only sampling is available on the service side. Frames are not transferred to the service. The client trains the model, gets embedding, makes an inference, on his side
445
$/month
/cloud pro
A data set of up to 100K frames
1 556
$/month
For an additional 100K frames
/on-premise
The custom model and the custom embedder are located in the service. All calculations are performed on the service side
For all
1 556
889
$/month
$/month
The service is entirely hosted on the client side (1 model and 1 embedder)
order a free call
By clicking the "Request a Call" button, you agree to the processing of your personal data in accordance with our Privacy Policy
GET IN TOUCH
order a free call
By clicking the "Request a Call" button, you agree to the processing of your personal data in accordance with our Privacy Policy
