Skip to content

VISUS Osnabrück derivative recovery

GazeForge freezes a reviewed recovery checkpoint for the historical University of Osnabrück multimedia-container conversion of the Kurzhals et al. VISUS benchmark. This checkpoint follows the VISUS source-resolution and 2021 supplementary-material recovery records. It materially improves the recoverable empirical structure while preserving the unresolved authority and rights gates around the original benchmark.

Result

A historical institutional derivative containing the 11 standard VISUS scenarios and 25 participant gaze tracks per scenario was structurally recovered and verified.

This is not the same claim as recovering a current authoritative copy of the original VISUS distribution.

The frozen record is:

validation/evidence/visus-source-recheck/
visus-osnabrueck-derivative-recovery-evidence-v1.json

Canonical evidence fingerprint:

78b538ad35e411fe9e7020d47c759d2d6d246c96887bf76da979e23f35335d32

Institutional source

The Osnabrück Institute of Cognitive Science page for its multimedia-container work explicitly lists the Kurzhals et al. [K] dataset — 11 videos among its converted eye-tracking datasets and offers ASS and USF versions. The page states that the converted datasets are provided for research purposes and requests citation of both the converter work and the corresponding original dataset.

The reviewed bundle endpoint is:

https://w3o.ikw.uni-osnabrueck.de/media/cv/mm_mkv/Kurzhals.zip

The legacy host currently presents an invalid TLS certificate chain. GazeForge's reconnaissance therefore records that TLS verification did not succeed and that retrieval required the runner's explicitly bounded insecure-TLS mode. That retrieval fact is never promoted into source authority.

The server returned a real ZIP object with:

  • remote size: 2,598,730,485 bytes;
  • ZIP central-directory entries: 26;
  • USF MKV entries: 13;
  • ASS MKV entries: 13;
  • standard USF scenarios: 11;
  • two additional USF polygon variants for K1 and K2;
  • two corresponding ASS polygon variants.

Only the ZIP metadata and selected members were inspected. Source bytes were not committed to GazeForge and were not uploaded as workflow artifacts.

Exact reconnaissance binding

The evidence record binds the successful temporary reconnaissance rather than relying on a narrative transcription:

  • branch: research/visus-osnabrueck-derivative-probe;
  • exact head: f4d614f67e519a90aa8326f562fac2be9772e825;
  • workflow: VISUS Osnabrueck derivative recovery probe;
  • run: 34504772669 (run number 6), conclusion success;
  • artifact: visus-osnabrueck-derivative-probe, ID 10163474159;
  • artifact digest: sha256:f4b40d146cc7cd17478da6bb2752ce7f80028f62bbf6356dd9ffc5ab6d048e76.

The artifact contains four compact JSON probe reports. Their raw byte counts and SHA-256 hashes are frozen in the evidence record. Recovered MKV or ZIP source bytes are absent from the artifact.

Why USF matters

Schöning et al., Visual Analytics of Gaze Data with Standard Multimedia Players (10.16910/jemr.10.5.4), describe the USF-based representation as encapsulating complete gaze metadata without loss, whereas the ASS conversion retains only selected metadata for visualization.

That publication-level statement makes USF the scientifically preferable recovery candidate. GazeForge nevertheless keeps converter_transform_fidelity_independently_proven_for_this_archive=false because the recovered USF files have not been compared byte-for-byte or field-for-field with an authoritative copy of the original VISUS exports.

The JEMR article's CC BY 4.0 license covers the article. It is not projected onto the converted VISUS dataset files.

Eleven-scenario structural recovery

The workflow reconstructed each of the 11 standard USF members independently from exact byte ranges of the 2.6 GB ZIP. For every member it verified the ZIP local-header metadata, decompressed size, CRC-32, compressed SHA-256, inflated SHA-256, Matroska/EBML signature, media stream properties, participant-track identity, and AOI-track structure. Each reconstructed member was deleted from the ephemeral runner after inspection.

All eleven scenarios passed the following common checks:

  • exactly 25 participant tracks with participant numbers 1..25;
  • group suffix distribution A=13, B=12;
  • participant payloads contain gaze, fixation, timestamp, and point metadata;
  • video resolution 1920 × 1080;
  • average frame rate 25/1 fps;
  • ZIP CRC verification;
  • valid Matroska EBML identity.

The standard USF set totals 1,288,160,237 compressed bytes and 1,305,649,112 inflated bytes.

Scenario USF member AOIs recovered
K1 01-car pursuit_usf.mkv Red Car; White Car
K2 02-turning car_usf.mkv Red Car
K3 03-dialog_usf.mkv Left Face; Right Face; Shirt
K4 04-thimblerig_usf.mkv Cup2; Cup1; Cup3
K5 05-memory_usf.mkv Cards
K6 06-UNO_usf.mkv Left Hand; Right Hand; Stack Covered; Stack Uncovered
K7 07-kite_usf.mkv Person; Kite
K8 08-case exchange_usf.mkv Persons; Textbox; Case; Suspects
K9 09-ball game_usf.mkv Ball; Player White; Player Red1; Player Red2; Player Red3
K10 10-bag search_usf.mkv Red Bag; Yellow Bag; Blue Bag; Red-White Bag; Personen; Brown Bag
K11 11-person search_usf.mkv Hooded; Red Shirt and Hat; Persons

Every scenario has its own immutable structural fingerprint in the evidence record. Each fingerprint binds its compressed/inflated file identities, video properties, exact participant roster and group counts, aggregate participant payload manifest, AOI titles, and aggregate AOI payload manifest.

What this establishes

This tranche establishes a substantially stronger derivative-source fact than the earlier small public fragments: a historical university-hosted conversion of all 11 standard VISUS scenarios is still recoverable, and its internal structure contains the expected 25-participant gaze corpus on every scenario. The recovered resolution, frame rate, participant cardinality, A/B suffix structure, and gaze metadata are consistent with the published benchmark.

This is sufficient to freeze the derivative recovery and structural identity of the inspected files. It is not sufficient to promote the original-source authority gate.

What remains unresolved

The following remain explicitly false:

  • current authoritative original VISUS copy recovered;
  • exact identity to the original VISUS files;
  • original VISUS dataset license resolved;
  • analysis-use rights resolved;
  • raw-source redistribution rights resolved;
  • insecure legacy-host retrieval as authority evidence;
  • identity of derivative AOI payloads to the original ViPER XML annotations;
  • independent human annotation streams verified;
  • VISUS source-audit authorization;
  • human-human or model-human validation authorization;
  • cross-dataset or native-60-Hz/GP3 validity authorization;
  • empirical Frozen Evidence authorization;
  • raw-source redistribution;
  • any new empirical performance claim.

The source page's phrase “provided for research purposes” is preserved as an important distribution-intent fact, but is not rewritten as a formal dataset license. Similarly, the converter publication's article license is not treated as a license for the converted dataset.

AOI identity caution

The USF files contain semantically named rectangular AOI tracks. Some recovered track titles are not textually identical to labels shown in the 2014 benchmark paper figures. Therefore structural recovery of AOI streams does not prove that the converter representation is a field-identical transformation of the original ViPER XML annotations.

This distinction matters for model-human validation: GazeForge will not silently substitute a derivative AOI stream for an authoritative original annotation without explicit review.

Participant-group caution

The participant tracks preserve P#A/P#B suffixes with 13 A and 12 B tracks per scenario. These suffixes are retained as provenance. They are not interpreted as different task conditions for every scenario; the original study used scenario-specific task differences only where documented.

Validation

The fail-closed validator is:

from gazeforge.visus_osnabrueck_derivative_recovery import (
    validate_visus_osnabrueck_derivative_recovery,
)

validate_visus_osnabrueck_derivative_recovery(
    "validation/evidence/visus-source-recheck/"
    "visus-osnabrueck-derivative-recovery-evidence-v1.json"
)

The validator revalidates both the existing authoritative-source recheck and the 2021 supplement-recovery checkpoint, hard-binds the exact reconnaissance run/artifact and all four report hashes, recomputes every scenario structural fingerprint, and rejects refingerprinted attempts to promote authority, rights, annotation independence, empirical validation, or redistribution.

Next source-resolution step

The next legitimate promotion route is now narrower and better defined:

  1. obtain an author/current-custodian-verified original VISUS copy or an authoritative redeposit tied to the 2014 benchmark;
  2. resolve analysis-use and redistribution rights for that exact source copy;
  3. compare its videos, gaze exports, and AOI annotations against the recovered Osnabrück USF derivative;
  4. only if that comparison and rights review succeed, construct the reviewed source-audit certificate and open empirical execution.

Until then, the Osnabrück recovery is frozen as strong derivative evidence, not as a substitute for the current authoritative source.