# OmniFall Repository Structure

## Directory Layout

```
parquet/                     Pre-built parquet files for all configs (used by load_dataset)
labels/                      CSV files containing temporal segment annotations
  caucafall.csv ... up_fall.csv   Staged dataset labels (7 columns)
  OOPS.csv                        OF-ItW labels (7 columns)
  of-syn.csv                      OF-Syn labels (19 columns: 7 core + 12 demographic)
splits/
  cs/                        Cross-subject splits (staged + OOPS)
  cv/                        Cross-view splits (staged + OOPS)
  syn/                       Synthetic splits (random, cross_age, cross_ethnicity, cross_bmi)
videos/metadata.csv          OF-Syn video-level metadata (12,000 videos)
data_files/
  omnifall-synthetic_av1.tar      OF-Syn video archive (12,000 AV1-encoded MP4s)
  syn_frame_wise_labels.tar.zst   OF-Syn frame-wise HDF5 labels
  oops_video_mapping.csv          Mapping from OOPS original filenames to OF-ItW names
prepare_oops_videos.py       Script to extract OOPS videos for OF-ItW
generate_parquet.py          Script to regenerate parquet files from labels + splits
```

## Label Format

Each label file in `labels/` contains temporal segment annotations:

```csv
path,label,start,end,subject,cam,dataset
path/to/clip,class_id,start_time,end_time,subject_id,camera_id,dataset_name
```

| Column | Type | Description |
|---|---|---|
| `path` | string | Relative path to the video (dataset-specific root) |
| `label` | int | Class ID (0-15), see [LABELS.md](LABELS.md) |
| `start` | float | Segment start time in seconds |
| `end` | float | Segment end time in seconds |
| `subject` | int | Subject ID (-1 for OF-ItW and OF-Syn) |
| `cam` | int | Camera view ID (-1 for OF-ItW and OF-Syn) |
| `dataset` | string | Source dataset name |

### OF-Syn Extended Labels

`labels/of-syn.csv` contains 12 additional demographic and scene metadata columns:

| Column | Values |
|---|---|
| `age_group` | toddlers_1_4, children_5_12, teenagers_13_17, young_adults_18_34, middle_aged_35_64, elderly_65_plus |
| `gender_presentation` | male, female |
| `monk_skin_tone` | mst1 through mst10 |
| `race_ethnicity_omb` | white, black, asian, hispanic_latino, aian, nhpi, mena |
| `bmi_band` | underweight, normal, overweight, obese |
| `height_band` | short, avg, tall |
| `environment_category` | indoor, outdoor |
| `camera_shot` | static_wide, static_medium_wide |
| `speed` | 24fps_rt, 25fps_rt, 30fps_rt, std_rt |
| `camera_elevation` | eye, low, high, top |
| `camera_azimuth` | front, rear, left, right |
| `camera_distance` | medium, far |

These fields are available in OF-Syn configs (`of-syn`, `of-syn-cross-*`, `labels-syn`, `metadata-syn`, `framewise-syn`). Cross-domain configs that mix sources use only the 7 core columns.

## Split Format

Split files in `splits/` list video paths included in each partition:

```csv
path
path/to/clip
```

Each video path maps to one or more temporal segments in the corresponding label file. The parquet configs contain the pre-merged result of joining split paths with labels.