Task 130 — Complete-Natural-Sequence Gate A Preview
Revised design evidence only. Each input/target pair below contains one complete publisher-defined natural sequence. The historical 25-frame preview is superseded. These are not canonical four-file samples, Gate B evidence, Final/training data, or model outputs.
Frozen proposal: 333 complete sequences containing all 36,214 paired frames, with zero fixed-window slicing and zero dropped tails.
Exact task-wide prompt:
Remove motion blur from this video and restore a sharp video. Preserve the original scene content, camera trajectory, subject motion, frame order, timing, framing, resolution, colors, textures, and illumination; do not stabilize, crop, recolor, retime, add, or remove anything.
Review all six complete pairs before attesting to the revised design.
Owner full-sequence checkpoint:6/6 PASS — On 2026-07-26, Huayi watched all six complete input/target pairs and read each repeated exact prompt and Gate A design facts block; no sample-specific exception was reported.
Review artifact: PR #184 · exact head cc3bbad7965a23b4651d66f0fb13c2be04b2ef7b
Privacy-sensitive public street/crowd; faces, signs, and vehicle identifiers may be visible.
Reviewer question
Does the complete sequence request only deblurring while preserving every person/object, crowd motion, camera path, framing, timing, and color?
Exact proposed prompt for this reference: Remove motion blur from this video and restore a sharp video. Preserve the original scene content, camera trajectory, subject motion, frame order, timing, framing, resolution, colors, textures, and illumination; do not stabilize, crop, recolor, retime, add, or remove anything.
Input reference — blur_gamma
Target reference — upstream sharp
Watch faces, limbs, vehicles, signs, and stone/building detail.
Verify continuous crowd and camera motion without stabilization or retiming.
Record sensitive faces, signs, or plates for the deferred privacy checkpoint.
Short-sequence boundary; strong spatially varying blur; privacy review remains applicable to real imagery.
Reviewer question
Are all 48 paired frames useful as one natural sample, with no slicing, tail dropping, padding, overlap, looping, or retiming?
Exact proposed prompt for this reference: Remove motion blur from this video and restore a sharp video. Preserve the original scene content, camera trajectory, subject motion, frame order, timing, framing, resolution, colors, textures, and illumination; do not stabilize, crop, recolor, retime, add, or remove anything.
Input reference — blur_gamma
Target reference — upstream sharp
Watch sandal straps and pavement edges during fast camera motion.
Verify the target introduces no crop, viewpoint shift, or color change.
Confirm the full 48-frame publisher sequence remains intact.
Does the complete mild-blur sequence remain useful without encouraging oversharpening or reconstruction beyond the paired target?
Exact proposed prompt for this reference: Remove motion blur from this video and restore a sharp video. Preserve the original scene content, camera trajectory, subject motion, frame order, timing, framing, resolution, colors, textures, and illumination; do not stabilize, crop, recolor, retime, add, or remove anything.
Input reference — upstream Blur/RGB
Target reference — upstream Sharp/RGB
Inspect grille, fence, rail, vehicle, and background branch detail.
Distinguish beam-splitter photometric/alignment differences from blur removal.
Check the entire pass for timing continuity and camera-path preservation.
Does “do not stabilize” correctly separate pixel deblurring from camera stabilization across the complete pan?
Exact proposed prompt for this reference: Remove motion blur from this video and restore a sharp video. Preserve the original scene content, camera trajectory, subject motion, frame order, timing, framing, resolution, colors, textures, and illumination; do not stabilize, crop, recolor, retime, add, or remove anything.
Input reference — upstream Blur/RGB
Target reference — upstream Sharp/RGB
Watch fences, signs, grids, and building edges across the full pan.
Verify the target preserves camera trajectory and event timing.
Check for optical mismatch, ringing, moiré, or geometry changes.
Does the prompt demand sharp pixels while preserving the complete camera trajectory and forbidding crop, recoloring, and invented detail?
Exact proposed prompt for this reference: Remove motion blur from this video and restore a sharp video. Preserve the original scene content, camera trajectory, subject motion, frame order, timing, framing, resolution, colors, textures, and illumination; do not stabilize, crop, recolor, retime, add, or remove anything.
Input reference — upstream Blur/RGB
Target reference — upstream Sharp/RGB
Inspect balcony rails and vertical edges through fast motion.
Check framing, sequence order, and camera trajectory from start to finish.
Separate exposure/alignment differences from the desired blur change.
Is the complete weak-difference sequence a useful positive example when static sections are subtle but the moving vehicle shows credible deblurring?
Exact proposed prompt for this reference: Remove motion blur from this video and restore a sharp video. Preserve the original scene content, camera trajectory, subject motion, frame order, timing, framing, resolution, colors, textures, and illumination; do not stabilize, crop, recolor, retime, add, or remove anything.
Input reference — upstream Blur/RGB
Target reference — upstream Sharp/RGB
Compare brick, foliage, wire, railing, and the passing vehicle.
Check exposure/optical differences between the two camera paths.
Decide whether the natural sequence should remain flagged rather than excluded.