Service overview
Any AI-driven product has it heart - Deep Neural Network, and product performs as well as DNN performs. Neural network performance is the result of training procedure - optimizing it internal weights based on training data. Our Service is not "just train the model and forget" but it includes bunch of steps to deliver the model that is ready for production usage:
- Data pre- and post-processing, training cycle and evaluating code development
- Search for the best architecture and architecture optimization
- Custom loss and metrics
- Model training
- Corner-cases analysis
- Integrating final model to your solution or model-as-service deployment
- Model support to continue train the model on new data
Models architectures
The list of model architectures we could train includes both classical architectures and SOTA including:
- ResNet-based networks
- MobileNet-based networks
- YOLO, SSD, Faster-RCNN, MaskRCNN
- Deeplabs and UNets
Use cases
- Face detection, tracking and recognition
- Object detection, tracking and recognition
- Semantic and instance segmentation
- Image generation using GAN-models
Supported framework
- Tensorflow 1.x and 2.x
- TFLite
- PyTorch
- Caffe and Caffe2
- Dlib
