[Data Quality] Inconsistent fullbright frame index - Frame 188 is not fullbright for multiple subjects

[fattypiggy]

Problem Summary

While processing the FaceOLAT dataset for foreground segmentation/matting tasks, I discovered that the fullbright frame index is inconsistent across different subjects and takes. Specifically, frame 188 (e.g., {expressionID}.000188.exr) - which I assumed to be the fullbright reference frame - does not contain full illumination for many subjects.

Expected Behavior

According to the dataset documentation, each take contains 350 lighting conditions. I expected a consistent frame index (e.g., frame 188) to represent the fullbright condition (all lights on) across all subjects, cameras, and takes.

Actual Behavior

The fullbright frame appears at different indices for different subjects. In some cases, the first 16 frames appear to be fullbright instead of frame 188.

Specific Example

Path	Expected	Actual
007/007/ID40007/Cam01/C006/ID40007.000188.exr	Fullbright frame	NOT fullbright (single light or partial illumination)
007/007/ID40007/Cam01/C006/ID40007.000001.exr to ID40007.000015.exr	Single light frames	Appear to be fullbright

Impact

This inconsistency affects downstream processing tasks such as:

1. Background matting/segmentation - Using incorrect frames produces wrong masks
2. Relighting pipelines - Incorrect fullbright reference affects light transport estimation
3. Any processing assuming consistent frame ordering

Questions

1. Is there a documented specification for which frame index corresponds to the fullbright condition?
2. Does the fullbright frame index vary by subject, expression, or capture session?
3. Is there metadata available that maps each frame to its lighting condition?
4. Could you provide guidance on how to reliably identify the fullbright frame for each take?