General description
Learning environments
Within the Mail.Ru Cloud Solutions platform, you can create a data scientists workspace.
The data scientist workspace is a virtual machine image that includes popular machine learning frameworks and tools.
Composition of the product
CS a: Ubuntu 18.04 "Machine Learning"
Pre-installed components:
- NVIDIA GPU drivers
- NVIDIA CUDA
- NVIDIA CUDA Deep Neural Network Librayr (cuDNN)
- NVIDIA Docker
- Anaconda Batch Manager
- C / C ++ development tools (gcc, g ++, clang, gdb, make, cmake, etc.)
- Version control systems (git, svn, mercurial)
Working with conda
Conda is a cross-platform package manager for Python, R, Ruby, Lua, Scala, Java, JavaScript, C / C ++, FORTRAN. Conda is an advanced counterpart to pip + virtualenv.
The conda repos contain all the popular machine learning tools and frameworks such as pandas , scikit-learn , Matplotlib , XGBoost , LightGBM , PyTorch , TensorFlow, and more.
Vision
Vision is a technology for recognizing faces, objects, processes based on machine learning and artificial neural networks. Vision will automate and improve the accuracy of complex visual inspection of varying complexity.
- 98% accuracy of face detection among a million
- TensorRT on inference, <10ms on photo on GPU
- 314 scene recognition classes, 25,000 object classes
Vision technologies are available through APIs that are constantly evolving. A list of them is available in this help center. Using the API, you can solve such cases as:
- Definition of scenes and objects
- Tracking people
- Celebrity recognition
- Recognizing text in images
- Vehicle detection
- Increase resolution
- Search for attractions
- Identification of a manufacturing defect
- Determination of car numbers
- Image moderation