jan-grzybek/aerial-film-restorer

The card points at the x2 that precedes this model in deployment

The card reads as a paper; the captions name the input once

The eight-channel model replaces the five-channel one; the card is measured per frame

Credit flows to the Powidok project

The checkpoints carry their hashes, and the card its thumbnail

The architecture credit carries the paper, and the card states the from-scratch lineage

Delete files __pycache__/* with huggingface_hub

Aerial Film Restorer v1: weights (CC-BY-4.0), inference machinery (MIT), measured-fidelity card

initial commit