Stockfilm Research
Changelog
Every version, correction, license grant and research sample, in date order. A correction produces a new version and an entry here; nothing is edited silently. Cite an edition by its permanent link and this page tells you what has happened to it since.
Entries
First published evaluation on the real-damage sample (Loop-Mamba, arXiv 2608.02346 v2)
The August 2026 revision of Loop-Mamba evaluates nine old-photo restoration methods on the 100-clip real-damage sample with BRISQUE, TOPIQ and MAN-IQA (Table 6) and shows qualitative comparisons (Figure 6). In v2 the in-text citation did not resolve and the reference list has no entry; the authors have been sent a BibTeX record for the next version.
Visual Memory Index aggregate data released under CC BY 4.0
Every edition's CSV downloads, release.json, figures and report text are now licensed under Creative Commons Attribution 4.0 International. Earlier text said formal open-data licensing was under review. The grant is retroactive to the 2026 edition. Footage, frames and catalog metadata records are not covered.
BibTeX, plain-text citation, citation badge and this changelog added
The report and media kit now carry a copyable BibTeX record and a resolvable citation, a small SVG badge with HTML and Markdown snippets for papers and repositories, and schema.org license and citation fields. A DOI is not yet issued; the permanent edition URL is the identifier until one exists.
Public loader and sample manifest repository
A small public GitHub repository holds a Python loader for the index's CSV files, the evaluation-only terms used for research samples, and a public manifest of a 100-clip real-damage evaluation set (clip IDs, dates, places and public clip pages; no media files).
Evaluation reference set for a film-restoration vendor
The 100-clip real-damage evaluation sample was re-wrapped for a commercial film restoration and digitization vendor under commercial evaluation terms. Same clip selection as the August research sample.
Dated-positive evaluation sample for a memory-modelling group
100 clips with confirmed shot years, prepared for a university group studying temporal grounding. Preparing it confirmed that every one of the archive's 217,560 restored assets had been published at some point; the archive holds no never-published footage.
Full-metadata sample in machine-learning formats for a law-school library
100 clips with complete catalog metadata as JSONL and CSV plus frames in Hugging Face imagefolder layout, prepared for a university library's digital-collections team.
Colour versus black-and-white evaluation sample
100 clips labelled colour or monochrome from chroma measurement of the masters, because catalog keywords proved unreliable for that attribute. Prepared for a colourisation research group.
European landmark evaluation sample
100 clips of identifiable European landmarks, prepared for a university group working on neural reconstruction from historical footage.
Real-damage evaluation sample (100 clips, six decades, 40 countries)
First external research sample: 100 digitized clips and losslessly extracted frames carrying real period damage (base scratches, dye fade, gate weave, splice damage, dust, emulsion wear), balanced across the 1930s to 1980s. Evaluation-only terms; no training use; citation requested.
Visual Memory Index 2026 v1.0: Initial release
Report stockfilm-vmi-2026-v1.0. Archive snapshot 2026-07-15, demand window 2026-02-27 to 2026-07-14, methodology v1.0, taxonomy v1.0.
What gets an entry
- A new edition or version of any research publication, with its report identifier.
- A correction to a published figure, with the old and new values in the release's own version log.
- A change to the data license or citation format.
- A research sample issued to an external group, described by institution type and purpose. Recipients are named only with their permission.
- Tooling that changes how the work can be cited or loaded.
Research samples are not public downloads. Groups that want one can start from the dataset page.