The NVIDIA Data Loading Library (DALI) is a portable, open-source software library designed to accelerate deep learning applications by decoding and augmenting images, videos, and speech. DALI mitigates bottlenecks by overlapping AI training and data pre-processing, providing a high-performance alternative to built-in data loaders and data iterators in popular deep learning frameworks.

Features

• Rapid Prototyping: Easy-to-use Python APIs and transparent scaling across multiple GPUs for quick iteration.
• GPU Acceleration: Accelerates training and inference for image, video, 3D volumes, and audio.
• Data Support: Supports multiple data formats including LMDB, TFRecord, COCO, JPEG, PNG, TIFF, JPEG2k, wav, flac, ogg, H.26, HEVC, and more.
• Custom Pipelines: Create custom pipelines with flexible graphs and add custom audio, image, and video processing operators.
• DALI Operators: Supports data loading operators for audio, video, and image processing.

Benefits

• Enhanced Performance: Reduces data access latency and training time, mitigating bottlenecks.
• Portability: Easily retarget data processing pipelines to TensorFlow, PyTorch, and MXNet.
• Flexibility: Use flexible graphs to create custom pipelines and add custom processing operators.
• Interoperability: Interoperable with various libraries, SDKs, and frameworks including Video Codec SDK, Video Processing Framework (VPF), Optical Flow SDK, TAO Toolkit, NVIDIA TensorRT, and Triton Inference Server.