TorchOK pipeline features
Extensive library of computer models: from classic ResNet to the latest Swin transformer;
The wide selection of ready-made datasets: it is only necessary to prepare your data in the required format (CSV file with annotations and paths to images);
Modern infrastructure. TorchOK runs on machines with CPU, GPU, and also on multiple computers with multiple GPUs. There is support for TPU;
Metrics for evaluating computer vision models in TensorBoard and MLflow: classification, segmentation, metrics for finding similar images, for face recognition;
The single interface for loading and unloading computer models;
Convenient "packaging": TorchOK can be run through the Conda environment and in the cloud — using Docker containerizer or in Safe Maker on Amazon Web Services.