random-sequence/flock-video-inconsistency

FLock Video Inconsistency

Short procedurally generated videos (default 320x240 at 15 fps, 6 to 10 s), each a continuous shot into which 0 to 4 known inconsistencies were injected, together with legitimate, unlabelled decoy events that look like edits but are not. Every clip comes with frame-exact labels. The data trains detectors for the FLock video_inconsistency task: given a video, list every inconsistency with its type, time span, confidence and (for two types) a bounding box.

• Suite version: video_inconsistency_v2; dataset build: video_inconsistency_hf_v2.
• License: CC-BY-4.0. All footage is synthetic.
• Access: public. Download it with huggingface_hub.snapshot_download(repo_type="dataset") (see Loading).

Loading

import json
from pathlib import Path
from huggingface_hub import snapshot_download

root = Path(snapshot_download("random-sequence/flock-video-inconsistency", repo_type="dataset"))
rows = [json.loads(line) for line in (root / "train" / "metadata.jsonl").read_text().splitlines()]
first = rows[0]
video_path = root / "train" / first["file_name"]   # mp4 (H.264, yuv420p)
print(first["difficulty"], first["issues"], first["decoys"])

The trainer sample kit reads a split directly: python train.py --data-dir <root>/train --out-dir runs/x (or --data-dir <root> to use train/ for training and validation/ for validation).

Files

Path	Content
train/metadata.jsonl, train/<clip_id>.mp4	training split
validation/metadata.jsonl, validation/<clip_id>.mp4	validation split (labels included)
dev_package/video_inconsistency_dev_package.zip	package for local validation with the validator
issue_types.json	canonical issue and decoy catalogue
stats.json	counts per split

Each metadata row: file_name, clip_id, fps, num_frames, width, height, duration, difficulty, source, crf (H.264 quality the clip was encoded at), issues and decoys. issues entries: type, start_time, end_time, start_frame, end_frame, bbox (spatial types only) and params (how the edit was made). decoys entries: type, start_time, end_time.

Time convention

Frame i covers [i / fps, (i + 1) / fps). A span over frames s..e inclusive has start_time = s / fps and end_time = (e + 1) / fps (end_frame is exclusive). A point event (dropped_frames) at the cut before frame k has start_time == end_time == k / fps and start_frame == end_frame == k. Bounding boxes are normalised [x0, y0, x1, y1] in [0, 1]; for a spatial issue the box is the union of the affected region over its whole time span.

The 10 issue types

Counts are train / validation.

Type	What to look for	Label	Count
frozen_frames	Motion stops: the same frame is held for a run of frames, then motion resumes with a jump.	span	1448 / 188
dropped_frames	A run of frames was cut out mid-shot, so objects and the camera jump forward instantly. Reported at the cut.	point event	1425 / 169
reversed_segment	A segment plays backwards: motion runs in reverse, then snaps forward.	span	1423 / 176
spliced_footage	Frames from an unrelated shot are inserted into the middle of a continuous shot.	span	1437 / 175
color_grade_jump	The colour grade / white balance changes abruptly for a segment and then changes back.	span	1364 / 172
exposure_flicker	One to three frames are much brighter or darker than their neighbours.	span	1452 / 175
mirrored_segment	A segment is horizontally flipped, so the scene's layout swaps sides.	span	1426 / 194
zoom_jump	A segment is abruptly punched in (cropped and scaled up), then snaps back to the original framing.	span	1428 / 181
inserted_object	A foreign object is pasted into the frame; it pops in and out abruptly and does not interact with the scene.	span + bbox	1443 / 208
blurred_region	A rectangular region is blurred or pixelated for a segment, as if retouched or censored.	span + bbox	1391 / 193

Decoys (hard negatives)

Decoys are real events in the shot that a detector must not report. They are listed in decoys, are never scored, and are never given to the detector at evaluation time. Use them as hard negatives.

Decoy	Description	Count (train / validation)
scene_cut	The shot changes to a different scene for good (an ordinary cut, not an inserted fragment).	614 / 90
exposure_drift	Brightness drifts smoothly up or down over a stretch (auto-exposure, a passing cloud).	625 / 89
white_balance_drift	The colour cast drifts smoothly over a stretch (auto white balance catching up).	664 / 72
smooth_zoom	The camera zooms in or out gradually (a normal push-in or pull-out).	662 / 82
object_enters	An object moves into the frame from outside, following the scene's motion.	679 / 89
object_exits	An object moves out of the frame, following the scene's motion.	693 / 94
object_stops	An object comes to rest while the rest of the scene keeps moving.	481 / 54
camera_stops	The camera pan slows to a halt and holds still (natural, gradual deceleration).	565 / 73
illumination_flicker		672 / 83
auto_exposure_step		686 / 89
auto_white_balance_step		681 / 76
fast_zoom		610 / 86
camera_direction_change		535 / 68
camera_speed_change		563 / 78

Difficulty tiers

Tier	Description	Clips (train / validation)
easy	1 edit per clip; long and strong, obvious to a casual viewer.	778 / 107
medium	1-2 edits per clip; moderate length and strength.	1924 / 228
hard	2-3 edits per clip; short and subtle.	2876 / 352
expert	2-4 edits per clip; the shortest and subtlest edits, together with the most decoys.	2422 / 313

Clips are also degraded after editing (sensor noise, optional blur or sharpening and down-up rescaling, then H.264 compression at a per-clip crf), so an edit must be detectable after degradation. Every edit remains visible to a careful human looking at the frames.

Split sizes

Split	Clips	Clean clips	Issues	Decoys	Duration	Size
train	8000	1611	14237	8730	17.77 h	1.43 GB
validation	1000	190	1831	1123	2.24 h	0.18 GB

Generation

• Generator: validator.modules.video_inconsistency.synthesis (generate_clip), seed 20260929; each clip's seed is a hash of (split, seed, index, attempt), so train and validation are disjoint from each other and from the dev package (seed 7).
• Resolution 320x240, 15 fps, 6 to 10 s per clip, fraction of clips edited from real footage: 0.
• Every mp4 was decoded back and verified to contain exactly num_frames frames.
• Rebuild: python -m validator.modules.video_inconsistency.build_hf_dataset --out-dir <dir> --seed 20260929.

Local validation with the dev package

dev_package/video_inconsistency_dev_package.zip is a validation package with 200 clips built with seed 7 by package.build_validation_package, in exactly the format the validator loads. Use it for local validation of a submission:

python environment_entrypoint.py video_inconsistency --local-validation \
    --hg-repo-id ./my_submission --validation-data-url ./video_inconsistency_dev_package.zip

It is disjoint from train and validation (separate seed namespace), so scores on it are honest.

Scoring

Detectors are scored with mean average precision (mAP): for every issue type, all of a detector's predictions (ranked by their confidence, at most 25 per clip) are matched to the ground truth at temporal-IoU thresholds 0.3, 0.4, 0.5, 0.6, 0.7, and the average precision is averaged over thresholds and over the issue types. Spatial types (inserted_object, blurred_region) additionally need a bounding-box IoU of at least 0.3 to count as a match. score = 0.75 * mean AP + 0.2 * localization + 0.05 * clip-level accuracy; clips are weighted by difficulty (easy 1, medium 1.5, hard 2, expert 2.5). Confidences should be honest probabilities, boundaries must be precise, and decoys (below) must not be reported.

The private evaluation set

The validators score submissions on a private evaluation set generated with a different, secret seed (and, ideally, different content). It shares the generator and the label semantics with this dataset but no clip. Do not expect a detector that memorises these clips to do well.