tao-train-action-recognition

To train an action recognition model with TAO, you use the TAO Core dataclass schemas to configure and select models, processing RGB, optical flow, or joint multi-stream inputs to classify temporal actions in video clips.

Yes, executing TAO action recognition workflows requires Docker and nvidia-container-toolkit to be installed in your environment to handle the containerized model training, evaluation, and inference processes.

Yes, you can export trained TAO action recognition models directly to ONNX format, enabling streamlined deployment for real-time inference on new video sequences.

TAO action recognition training supports RGB, optical flow, and joint multi-stream input types to accurately classify temporal actions within video data.

You run real-time inference on video sequences by deploying your trained and evaluated TAO action recognition model, which processes the video data to classify temporal actions.

Yes, TAO utilizes TAO Core dataclass schemas to provide automated configuration and model selection, streamlining the setup process for video action classification tasks.