Pure Rust video action classification engine based on ViViT. Classifies video clips into 400 action categories from Kinetics-400. No Python runtime, no CUDA, no external dependencies.

Try: https://huggingface.co/qoranet/QORA-Vision-Video

Classify from frame directory (fast, from binary)

qora-vision.exe vivit --load model.qora-vision --frames ./my_frames/

Classify from video file (requires ffmpeg)

qora-vision.exe vivit --load model.qora-vision --video clip.mp4

Load from safetensors (slow, first time)

qora-vision.exe vivit --frames ./my_frames/ --model-path ../ViViT/

Save binary for fast loading

qora-vision.exe vivit --model-path ../ViViT/ --save model.qora-vision