aidkit is the data solution for better training and performance

Getting the data you need to train your perception model to handle every condition your ODD can throw at is challenging. So if you can't change the weather to get the data you need, create the data you need with aidkit.

aidkit takes your existing images and augments them with the conditions and intensities you specify, allowing better model performance and allocation of rare real-world data for testing that proves that performance.

No matter the model, scene, or ODD, aidkit ’til you make it.

aidkit

Balance reality and simulation with augmented data for training better models

aidkit is the toolkit for generating realistic data augmentations for training your ADAS/ADS perception models and enhancing their coverage of your ODD.

Without the right data no amount of training will enhance your model's accuracy, but getting the right data is not easy: until now.

aidkit generates altered scences with whatever conditions you need, ensuring your training dataset represents the whole reality your model needs to handle.

aidkit augmentations are not only tested with statistical validation to ensure they're mathmatically real, but built to pass human validitation by looking real.

Augmentation

Rapid augmented data generation

aidkit generates a wide range of altered images with the precise parameters you specify in order to fill gaps in your data. Obtain a better data distribution or fill in edge and corner cases. Data generation is scalable so you don't need to delay development to wait on data campaigns.

Training

Accurate representation of reality for enhanced accuracy

Train with confidence that your augmented data is as good as the real thing. With method validity tests and old-fashioned visual inspection, you'll be able to trust aidkit's augmented data. And if you need more, our team is ready to build your customized augmentations.

Performance

Greater coverage and lower failure rates

Experience performance improvements of circa 5 to 10%. Enhanced performance means your model has better coverage of all the possible conditions in your ODD and a lower failure rate, one falling within an acceptable risk level. This means your ADAS/ADS will perform more safely in more situations.

aidkit: trust in your AI begins with Training data you can trust

Simplify and master perception models

By augmenting data with perturbation-based image operations, aidkit improves the training of perception models. It allows the inclusion of new data capturing various conditions that are possible in an ODD but not easily captured in fleet data. Training with augmented data allosw you to save scarce real world data for validation and testing, while being confident the training data quality was high enough to ensure good ex post performance.

Data augmentation lies between reality and simulation. Because it uses real data as basis upon which it adds augmentations, it lies closer to reality than fully simulated data. This allows easier testing to ensure the augmented data accurately represents reality and can be reliabily used in ML model training. aidkit and neurocat take care of all this for you, so you can focus on getting your model into the camera, then the car, and finally on the road.