diff --git a/.claude/settings.local.json b/.claude/settings.local.json index ba20b998716a44cbaba11de6efaf71b429a1a4a4..e5d63cfed37b9b1499a28b72e78d321ccf2ccbcd 100644 --- a/.claude/settings.local.json +++ b/.claude/settings.local.json @@ -1,14 +1,7 @@ { "permissions": { "allow": [ - "Read(//home/dschneider/workspace/PROJECTS/omnifall/hf/**)", - "Read(//lsdf/users/dschneider-kf3609/workspace/PROJECTS/omnifall/hf/**)", - "Read(//lsdf/users/dschneider-kf3609/workspace/PROJECTS/wanfall/labels/**)", - "Bash(source ~/miniconda3/bin/activate ccode)", - "Bash(source:*)", - "WebSearch" - ], - "deny": [], - "ask": [] + "Bash(/home/dschneider/miniconda3_cvhci/envs/falldet/bin/python:*)" + ] } } diff --git a/.gitignore b/.gitignore index c8e89a39a9f02f0aa873782ddce11d6c6cbfefc5..fdef7a9555f5a83e06e151f02092e454725cc2ef 100644 --- a/.gitignore +++ b/.gitignore @@ -1,8 +1,4 @@ -CLAUDE.md -create_demographic_plots.py -create_splits.py -export_via_to_csv.py -extract_jsonl_metadata.py -test_wanfall_builder.py -.claude -test_framewise_complete.py +convert_oops_via_to_csv.py +# Symlink for local testing (HF derives dataset name from directory name) +hf.py +__pycache__ diff --git a/README.md b/README.md index d56cfe4d2d7f99984b9e4b37c42e24bed49d0669..73fe661a1e2734e5b5719945826b01700e1b51d6 100644 --- a/README.md +++ b/README.md @@ -5,468 +5,363 @@ task_categories: language: - en tags: -- synthetic -- activity-recognition +- medical - fall-detection -pretty_name: 'WanFall: A Synthetic Activity Recognition Dataset' +- activity-recognition +- synthetic +pretty_name: 'OmniFall: A Unified Benchmark for Staged-to-Wild Fall Detection' size_categories: - 10K + + arXiv + +
+ + arXiv + -**Motion Types:** -- **Dynamic** (0-3, 5, 7, 9-10, 12, 14-15): Transitions and movements -- **Static** (2, 4, 6, 8, 11, 13): Stationary postures +# OmniFall: A Unified Benchmark for Staged-to-Wild Fall Detection -## Data Format +This repository contains annotations, splits, and metadata for OmniFall, a comprehensive benchmark with three components: -### CSV Columns (19 fields) +- **OF-Staged (OF-Sta):** 8 public staged fall detection datasets (~14h single-view, 112 subjects, 31 camera views) +- **OF-In-the-Wild (OF-ItW):** Curated genuine accident videos from OOPS (~2.7h, 818 videos) +- **OF-Synthetic (OF-Syn):** 12,000 synthetic videos generated with Wan 2.2 (~17h) with controlled demographic diversity -```python -# Core annotation fields -path # Video path (e.g., "fall/fall_ch_001") -label # Activity class ID (0-15) -start # Segment start time (seconds) -end # Segment end time (seconds) -subject # -1 (synthetic data) -cam # -1 (single view) -dataset # "wanfall" - -# Demographic metadata (6 fields) -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-mst10 (Monk Skin Tone scale) -race_ethnicity_omb # white, black, asian, hispanic_latino, aian, nhpi, mena (OMB categories) -bmi_band # underweight, normal, overweight, obese -height_band # short, avg, tall - -# Scene metadata (6 fields) -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 -``` +All components share a sixteen-class activity taxonomy. Staged datasets use the core ten classes (0-9), while OF-ItW and OF-Syn use the full sixteen classes (0-15). -**References:** -- [Monk Skin Tone Scale](https://skintone.google/the-scale) - 10-point inclusive skin tone representation -- [OMB Race/Ethnicity Standards](https://www.census.gov/newsroom/blogs/random-samplings/2024/04/updates-race-ethnicity-standards.html) +[OmniFall: A Unified Staged-to-Wild Benchmark for Human Fall Detection](https://arxiv.org/abs/2505.19889) -## Split Configurations +Also have a look for additional information on our project page: -### 1. Random Split (80/10/10) +[OmniFall Project Page](https://simplexsigil.github.io/omnifall/) -Standard baseline with random video assignment (seed 42). +## Datasets -| Split | Videos | Segments | -|-------|--------|----------| -| Train | 9,600 | 15,344 | -| Val | 1,200 | 1,956 | -| Test | 1,200 | 1,928 | +### OF-Staged (8 public datasets) -```python -dataset = load_dataset("simplexsigil2/wanfall", "random") -``` - -### 2. Cross-Age Split - -Evaluates generalization across age groups. Train on adults, test on children and elderly. +1. **[CMDFall](https://www.mica.edu.vn/perso/Tran-Thi-Thanh-Hai/CMDFALL.html)** (7h 25m single view) - 50 subjects, 7 synchronized views +2. **[UP Fall](https://sites.google.com/up.edu.mx/har-up/)** (4h 35m) - 17 subjects, 2 synchronized views +3. **[Le2i](https://search-data.ubfc.fr/imvia/FR-13002091000019-2024-04-09_Fall-Detection-Dataset.html)** (47m) - 9 subjects, 6 different rooms +4. **[GMDCSA24](https://github.com/ekramalam/GMDCSA24-A-Dataset-for-Human-Fall-Detection-in-Videos)** (21m) - 4 subjects, 3 rooms +5. **[CAUCAFall](https://data.mendeley.com/datasets/7w7fccy7ky/4)** (16m) - 10 subjects, 1 room +6. **[EDF](https://doi.org/10.5281/zenodo.15494102)** (13m) - 5 subjects, 2 views synchronized +7. **[OCCU](https://doi.org/10.5281/zenodo.15494102)** (14m) - 5 subjects, 2 views not synchronized +8. **[MCFD](https://www.iro.umontreal.ca/~labimage/Dataset/)** (12m) - 1 subject, 8 views -| Split | Videos | Age Groups | -|-------|--------|------------| -| **Train** | 4,000 | `young_adults_18_34` (2,000)
`middle_aged_35_64` (2,000) | -| **Val** | 2,000 | `teenagers_13_17` (2,000) | -| **Test** | 6,000 | `children_5_12` (2,000)
`toddlers_1_4` (2,000)
`elderly_65_plus` (2,000) | +### OF-ItW (In-the-Wild) -```python -dataset = load_dataset("simplexsigil2/wanfall", "cross_age") -``` +9. **[OOPS-Fall](https://oops.cs.columbia.edu/data/)** (2h 39m) - Curated subset of genuine fall accidents from the OOPS dataset, strong variation in subjects and views. -### 3. Cross-Ethnicity Split +### OF-Syn (Synthetic) -Evaluates generalization across racial/ethnic groups with maximum phenotypic distance. Train on White/Asian/Hispanic, test on Black/MENA/NHPI. +10. **OF-Syn** (16h 53m) - 12,000 synthetic videos generated with Wan 2.2 video diffusion model. Features controlled demographic diversity across age (6 groups), ethnicity (7 OMB categories), body type (4 BMI bands), gender, and environments. Labels include 19 columns: 7 core annotation fields plus 12 demographic and scene metadata fields. -| Split | Videos | Ethnicities | -|-------|--------|-------------| -| **Train** | 5,178 | `white` (1,709)
`asian` (1,691)
`hispanic_latino` (1,778) | -| **Val** | 1,741 | `aian` (1,741) | -| **Test** | 5,081 | `black` (1,684)
`mena` (1,680)
`nhpi` (1,717) | +## Structure -```python -dataset = load_dataset("simplexsigil2/wanfall", "cross_ethnicity") -``` +The repository is organized as follows: -### 4. Cross-BMI Split +- `omnifall.py` - Custom HuggingFace dataset builder (handles all configs) +- `labels/` - CSV files containing temporal segment annotations + - Staged/OOPS labels: 7 columns (`path, label, start, end, subject, cam, dataset`) + - OF-Syn labels: 19 columns (7 core + 12 demographic/scene metadata) +- `splits/` - Train/validation/test splits + - `splits/cs/` - Cross-subject splits (staged + OOPS) + - `splits/cv/` - Cross-view splits (staged + OOPS) + - `splits/syn/` - Synthetic splits (random, cross_age, cross_ethnicity, cross_bmi) +- `videos/metadata.csv` - OF-Syn video-level metadata (12,000 videos) +- `data_files/syn_frame_wise_labels.tar.zst` - OF-Syn frame-wise HDF5 labels -Evaluates generalization across body types. Train on normal/underweight, test on obese. +### Label Format -| Split | Videos | BMI Bands | -|-------|--------|-----------| -| **Train** | 6,066 | `normal` (3,040)
`underweight` (3,026) | -| **Val** | 2,962 | `overweight` (2,962) | -| **Test** | 2,972 | `obese` (2,972) | +Each label file in the `labels/` directory follows this format: -```python -dataset = load_dataset("simplexsigil2/wanfall", "cross_bmi") ``` - -**Note:** All cross-demographic splits contain the same videos, just organized differently. Total unique videos: 12,000. - -## Usage - -The dataset provides flexible loading options depending on your use case. The key distinction is between **segment-level** and **video-level** samples. - -### Loading Modes Overview - -| Mode | Sample Unit | Has start/end? | Has frame_labels? | Random Split Train Size | -|------|-------------|----------------|-------------------|------------------------| -| **Temporal Segments** | Segment | ✅ Yes | ❌ No | 15,344 segments (9,600 videos) | -| **Frame-Wise Labels** | Video | ❌ No | ✅ Yes (81 labels) | 9,600 videos | - -### 1. Temporal Segments (Default) - -Load temporal segment annotations where **each sample is a segment** with start/end times. Multiple segments can come from the same video. - -```python -dataset = load_dataset("simplexsigil2/wanfall", "random") - -# Each example is a SEGMENT (not a video) -example = dataset['train'][0] -print(example['path']) # "fall/fall_ch_001" -print(example['label']) # 1 (activity class ID) -print(example['start']) # 0.0 (start time in seconds) -print(example['end']) # 1.006 (end time in seconds) -print(example['age_group']) # Demographic metadata - -# Dataset contains multiple segments per video -print(f"Total segments in train: {len(dataset['train'])}") # 15,344 -print(f"Unique videos: {len(set([ex['path'] for ex in dataset['train']]))}") # 9,600 +path,label,start,end,subject,cam,dataset +path/to/clip,class_id,start_time,end_time,subject_id,camera_id,dataset_name ``` -**Use case:** Training models on activity classification where you want to extract and process only the relevant video segment for each activity. - -### 2. Frame-Wise Labels - -Load dense frame-level labels where **each sample is a video** with 81 frame labels. Each video appears exactly once. - -```python -dataset = load_dataset("simplexsigil2/wanfall", "random", framewise=True) - -# Each example is a VIDEO (not a segment) -example = dataset['train'][0] -print(example['path']) # "fall/fall_ch_001" -print(example['frame_labels']) # [1, 1, 1, ..., 11, 11] (81 labels) -print(len(example['frame_labels'])) # 81 frames -print(example['age_group']) # Demographic metadata included +Where: +- `path`: Relative path to the video, given the respective dataset root. +- `label`: Class ID (0-15) corresponding to one of sixteen activity classes: + - 0: walk + - 1: fall + - 2: fallen + - 3: sit_down + - 4: sitting + - 5: lie_down + - 6: lying + - 7: stand_up + - 8: standing + - 9: other + - 10: kneel_down + - 11: kneeling + - 12: squat_down + - 13: squatting + - 14: crawl + - 15: jump +- `start`: Start time of the segment (in seconds) +- `end`: End time of the segment (in seconds) +- `subject`: Subject ID +- `cam`: Camera view ID +- `dataset`: Name of the dataset + +**Note on class usage:** Staged datasets (OF-Sta) use only the core ten classes (0-9). OF-ItW and OF-Syn use the full sixteen-class taxonomy. The extended classes (10-15) are infrequent and can be treated as "other" (class 9) for compatibility. + +Cam and subject IDs are -1 for OF-ItW and OF-Syn, since these datasets have no consistent camera views or subjects across videos. + +**Note on OF-Syn labels:** The `labels/of-syn.csv` file contains 19 columns: the 7 core fields above plus 12 demographic and scene metadata fields (age_group, gender_presentation, monk_skin_tone, race_ethnicity_omb, bmi_band, height_band, environment_category, camera_shot, speed, camera_elevation, camera_azimuth, camera_distance). These additional fields are only available when loading OF-Syn configs. + +### Split Format + +Split files in the `splits/` directory list the video segments included in each partition. You can use the split paths to filter the label data.: -# Dataset contains one sample per video -print(f"Total videos in train: {len(dataset['train'])}") # 9,600 videos ``` - -**Use case:** Training sequence models (e.g., temporal action segmentation) that process entire videos and predict frame-level labels. - -**Key features:** -- Works with all split configs: Add `framewise=True` to any split -- Efficient: 348KB compressed archive, automatically cached -- Complete metadata: All demographic attributes included - -### 3. Additional Configurations - -```python -# All segments without train/val/test splits -dataset = load_dataset("simplexsigil2/wanfall", "labels") # 19,228 segments - -# Video metadata only (no labels) -dataset = load_dataset("simplexsigil2/wanfall", "metadata") # 12,000 videos - -# Paths only (minimal memory footprint) -dataset = load_dataset("simplexsigil2/wanfall", "random", paths_only=True) +path +path/to/clip ``` -### Practical Examples +## Evaluation Protocols -#### Label Conversion +All configurations are defined in the `omnifall.py` dataset builder and loaded via `load_dataset("simplexsigil2/omnifall", "")`. -Labels are stored as integers (0-15) but can be converted to strings: +### Labels (no train/val/test splits) +- `labels` (default): All staged + OOPS labels (52k segments, 7 columns) +- `labels-syn`: OF-Syn labels with demographic metadata (19k segments, 19 columns) +- `metadata-syn`: OF-Syn video-level metadata (12k videos) +- `framewise-syn`: OF-Syn frame-wise HDF5 labels (81 labels per video) -```python -dataset = load_dataset("simplexsigil2/wanfall", "random") -label_feature = dataset['train'].features['label'] +### OF-Staged Configs +- `of-sta-cs`: 8 staged datasets, cross-subject splits +- `of-sta-cv`: 8 staged datasets, cross-view splits -# Convert integer to string -label_name = label_feature.int2str(1) # "fall" +### OF-ItW Config +- `of-itw`: OOPS-Fall in-the-wild genuine accidents -# Convert string to integer -label_id = label_feature.str2int("walk") # 0 +### OF-Syn Configs +- `of-syn`: Fixed randomized 80/10/10 split +- `of-syn-cross-age`: Cross-age split (train: adults, test: children/elderly) +- `of-syn-cross-ethnicity`: Cross-ethnicity split +- `of-syn-cross-bmi`: Cross-BMI split (train: normal/underweight, test: obese) -# Access all label names -all_labels = label_feature.names # ['walk', 'fall', 'fallen', ...] -``` +### Cross-Domain Evaluation +- `of-sta-itw-cs`: Train/val on staged CS, test on OOPS +- `of-sta-itw-cv`: Train/val on staged CV, test on OOPS +- `of-syn-itw`: Train/val on OF-Syn random, test on OOPS -#### Filter by Demographics +### Aggregate Configs (staged + OOPS combined) +- `cs`: Cross-subject splits for all staged + OOPS +- `cv`: Cross-view splits for all staged + OOPS -```python -dataset = load_dataset("simplexsigil2/wanfall", "labels") -segments = dataset['train'] - -# Filter elderly fall segments -elderly_falls = [ - ex for ex in segments - if ex['age_group'] == 'elderly_65_plus' and ex['label'] == 1 -] -print(f"Found {len(elderly_falls)} elderly fall segments") - -# Filter by multiple demographics -indoor_male_falls = [ - ex for ex in segments - if ex['environment_category'] == 'indoor' - and ex['gender_presentation'] == 'male' - and ex['label'] == 1 -] -``` +### Individual Dataset Configs +- `caucafall`, `cmdfall`, `edf`, `gmdcsa24`, `le2i`, `mcfd`, `occu`, `up_fall`: Individual datasets with cross-subject splits -#### Cross-Demographic Evaluation +### Deprecated Config Names -```python -# Train on young adults, test on children and elderly -cross_age = load_dataset("simplexsigil2/wanfall", "cross_age", framewise=True) - -# Train contains only: young_adults_18_34, middle_aged_35_64 -for example in cross_age['train'][:5]: - print(f"Train video: {example['path']}, age: {example['age_group']}") +The following old config names still work but emit a deprecation warning: -# Test contains: children_5_12, toddlers_1_4, elderly_65_plus -for example in cross_age['test'][:5]: - print(f"Test video: {example['path']}, age: {example['age_group']}") -``` +| Old Name | Use Instead | +|---|---| +| `cs-staged` | `of-sta-cs` | +| `cv-staged` | `of-sta-cv` | +| `cs-staged-wild` | `of-sta-itw-cs` | +| `cv-staged-wild` | `of-sta-itw-cv` | +| `OOPS` | `of-itw` | -#### Training Loop Example +## Examples ```python from datasets import load_dataset -import torch - -# Load dataset with frame-wise labels -dataset = load_dataset("simplexsigil2/wanfall", "random", framewise=True) - -for epoch in range(num_epochs): - for example in dataset['train']: - video_path = example['path'] - frame_labels = torch.tensor(example['frame_labels']) # (81,) - # Load video frames (user must implement) - # frames = load_video(video_root / f"{video_path}.mp4") # (81, H, W, 3) - - # Forward pass - # outputs = model(frames) - # loss = criterion(outputs, frame_labels) - # loss.backward() +# Load the three main components +staged = load_dataset("simplexsigil2/omnifall", "of-sta-cs") # 8 staged datasets +itw = load_dataset("simplexsigil2/omnifall", "of-itw") # OOPS in-the-wild +syn = load_dataset("simplexsigil2/omnifall", "of-syn") # OF-Syn synthetic + +# Each returns train/validation/test splits with labels already merged +print(f"Staged train: {len(staged['train'])} segments") +print(f"ItW test: {len(itw['test'])} segments") +print(f"Syn train: {len(syn['train'])} segments") + +# Synthetic data includes demographic metadata (19 columns) +example = syn['train'][0] +print(f"Path: {example['path']}, Label: {example['label']}") +print(f"Age: {example['age_group']}, Ethnicity: {example['race_ethnicity_omb']}") + +# Cross-domain: train on staged, test on wild +cross = load_dataset("simplexsigil2/omnifall", "of-sta-itw-cs") +print(f"Train on staged: {len(cross['train'])} segments") +print(f"Test on wild: {len(cross['test'])} segments") + +# All labels without splits (for custom splitting) +labels = load_dataset("simplexsigil2/omnifall", "labels")["train"] +syn_labels = load_dataset("simplexsigil2/omnifall", "labels-syn")["train"] + +# Synthetic frame-wise labels (81 labels per video) +framewise = load_dataset("simplexsigil2/omnifall", "of-syn", framewise=True) +print(f"Frame labels: {framewise['train'][0]['frame_labels'][:5]}...") # [1, 1, 1, ...] ``` -## Annotation Guidelines - -### Temporal Precision +## Label definitions -Annotations use sub-second accuracy with decimal timestamps (e.g., `start: 0.0, end: 1.006`). Most frames in videos are labeled, with minimal gaps between activities. +In this section we provide additional information about the labelling process to provide as much transparency as possible. -### Activity Sequences +The benchmark uses a sixteen-class taxonomy. Staged datasets use classes 0-9, while OOPS-Fall uses the full range 0-15. -Videos contain natural transitions between activities. Common sequences include: +### Core Classes (0-9, all datasets) +- **`0|walk`** - Move around, including jogging and running and "drunk walking", but only if it is not part of some special exercise like pulling your knees up. Not when pushing a large object like a chair, but included carrying something small like an apple. +- **`1|fall`** - The act of falling (from any previous state). Includes falling on a bed, if the process is not a controlled lying down with arms as support. +- **`2|fallen`** - Being on the ground or a mattress after a fall. +- **`3|sit_down`** - Sitting down on bed or chair or ground. +- **`4|sitting`** - Sitting on bed or chair or ground. +- **`5|lie_down`** - Lying down intentionally (in contrast to a fall) on ground or bed. +- **`6|lying`** - Being in a lying position (in bed or on the ground) after intentionally getting into that position. +- **`7|stand_up`** - Standing up from a fallen state, from lying or sitting. Includes getting from lying position into sitting position. +- **`8|standing`** - Standing around without walking. +- **`9|other`** - Any other activity, including e.g. walking while pushing an object like a chair. + +### Extended Classes (10-15, OF-ItW and OF-Syn) + +These classes capture additional activities observed in genuine accident and synthetic videos. They are infrequent and do not occur in staged datasets. + +- **`10|kneel_down`** - Transitioning from standing or another posture to a kneeling position. +- **`11|kneeling`** - Being in a kneeling position (static posture). +- **`12|squat_down`** - Transitioning to a squatting position. +- **`13|squatting`** - Being in a squatting position (static posture). +- **`14|crawl`** - Crawling on hands and knees or similar locomotion on the ground. +- **`15|jump`** - Jumping action, including vertical jumps and jumps from elevated positions. + +## Motion Types + +There are two types of motions, **dynamic** ones like `walk` or `stand_up` and **static** ones like `fallen`, `sitting`, `lying`. + +**Generally we annotate dynamic motions as soon as the first frame appears which belongs to that action**. + +Let's say we see a person `walk`, then `fall`. The first frame which indicates a motion which does seem to be different to `walk` is the start of `fall`. Sometimes it is necessary to have a look at a person walking to learn when the motion begins to change to something else. + +**For static motions the label begins with the first frame where the person comes to a resting state.** For `sit_down`, the label ends when the person reaches a state where it is no longer adjusting its body position but comes to a rest. `fall` ends when the person is no longer moving caused by the inertia of the fall. The following `fallen` might contain movement on the ground, but no movement which belongs to `fall` or `stand_up`. + +## Label Sequences + +There are some natural sequences of labels like `fall`, `fallen` and `stand_up`. However, it is not always the case that these appear together. Sometimes the person might directly stand up again without any time at rest on the ground, in this case there is no `fallen` segment. Likewise sometimes there is no sitting segment after sitting down (e.g. if the person immediately stands up again). + +Lying down can be on a bed or on the ground, it is intentional in contrast to `fall`. There are falls which follow `sit_down` or `lie_down` if it is from a chair or from a bed. + +When a person is lying in a bed and getting up to sit in the bed we label this as `stand_up`, even if the person is still sitting in the bed. A sequence could then be `lying`, `stand_up`, `sitting`, `stand_up` to describe a person which first lies down, then gets into a sitting position, waits a little, then gets fully up. + +Sometimes it is not perfectly clear if it is `sit_down` followed by `lie_down` or simply `lie_down`. This depends on there being a moment of rest or not. If the person spends a short amount of time in the sitting position it is the former, if the person directly goes from `standing` over `sit_down` to `lie_down` without rest it is only labeled `lie_down`. Similar thoughts apply to `stand_up`. + +In the video at the bottom of this page we show how we leveraged VGG VIA to perform the annotations. Note, that we pre-load the original dataset labels as visual aid, but relabel the whole video with our label definitions. In the video below, CMDFall is shown, which already provides relatively detailed labels, the original labels of other datasets are more sparse. Additionally, even CMDFall does not label all frames, but only specific segments, while our labels cover mostly every frame. + +The blurred out regions were added by us in post-processing to protect the subjects privacy on this page, they are not part of the original videos. + +## Citation + +If you use OmniFall in your research, please cite our paper (will be updated soon) as well as all sub-dataset papers: + +```bibtex +@misc{omnifall, + title={OmniFall: From Staged Through Synthetic to Wild, A Unified Multi-Domain Dataset for Robust Fall Detection}, + author={David Schneider and Zdravko Marinov and Rafael Baur and Zeyun Zhong and Rodi Düger and Rainer Stiefelhagen}, + year={2025}, + eprint={2505.19889}, + archivePrefix={arXiv}, + primaryClass={cs.CV}, + url={https://arxiv.org/abs/2505.19889}, +}, + +@inproceedings{omnifall_cmdfall, + title={A multi-modal multi-view dataset for human fall analysis and preliminary investigation on modality}, + author={Tran, Thanh-Hai and Le, Thi-Lan and Pham, Dinh-Tan and Hoang, Van-Nam and Khong, Van-Minh and Tran, Quoc-Toan and Nguyen, Thai-Son and Pham, Cuong}, + booktitle={2018 24th International Conference on Pattern Recognition (ICPR)}, + pages={1947--1952}, + year={2018}, + organization={IEEE} +}, + +@article{omnifall_up-fall, + title={UP-fall detection dataset: A multimodal approach}, + author={Mart{\'\i}nez-Villase{\~n}or, Lourdes and Ponce, Hiram and Brieva, Jorge and Moya-Albor, Ernesto and N{\'u}{\~n}ez-Mart{\'\i}nez, Jos{\'e} and Pe{\~n}afort-Asturiano, Carlos}, + journal={Sensors}, + volume={19}, + number={9}, + pages={1988}, + year={2019}, + publisher={MDPI} +}, + +@article{omnifall_le2i, + title={Optimized spatio-temporal descriptors for real-time fall detection: comparison of support vector machine and Adaboost-based classification}, + author={Charfi, Imen and Miteran, Johel and Dubois, Julien and Atri, Mohamed and Tourki, Rached}, + journal={Journal of Electronic Imaging}, + volume={22}, + number={4}, + pages={041106--041106}, + year={2013}, + publisher={Society of Photo-Optical Instrumentation Engineers} +}, + +@article{omnifall_gmdcsa, + title={GMDCSA-24: A dataset for human fall detection in videos}, + author={Alam, Ekram and Sufian, Abu and Dutta, Paramartha and Leo, Marco and Hameed, Ibrahim A}, + journal={Data in Brief}, + volume={57}, + pages={110892}, + year={2024}, + publisher={Elsevier} +}, + +@article{omnifall_cauca, + title={Dataset CAUCAFall}, + author={Eraso, Jose Camilo and Mu{\~n}oz, Elena and Mu{\~n}oz, Mariela and Pinto, Jesus}, + journal={Mendeley Data}, + volume={4}, + year={2022} +}, + +@inproceedings{omnifall_edf_occu, + title={Evaluating depth-based computer vision methods for fall detection under occlusions}, + author={Zhang, Zhong and Conly, Christopher and Athitsos, Vassilis}, + booktitle={International symposium on visual computing}, + pages={196--207}, + year={2014}, + organization={Springer} +}, + +@article{omnifall_mcfd, + title={Multiple cameras fall dataset}, + author={Auvinet, Edouard and Rougier, Caroline and Meunier, Jean and St-Arnaud, Alain and Rousseau, Jacqueline}, + journal={DIRO-Universit{\'e} de Montr{\'e}al, Tech. Rep}, + volume={1350}, + pages={24}, + year={2010} +}, + +@inproceedings{omnifall_oops, + title={Oops! predicting unintentional action in video}, + author={Epstein, Dave and Chen, Boyuan and Vondrick, Carl}, + booktitle={Proceedings of the IEEE/CVF conference on computer vision and pattern recognition}, + pages={919--929}, + year={2020} +} ``` -walk → fall → fallen → stand_up -walk → sit_down → sitting → stand_up -walk → lie_down → lying → stand_up -standing → squat_down → squatting → stand_up -``` - -Not all transitions include static states. For example, a person might `stand_up` immediately after falling without a `fallen` state. - -### Motion Types - -**Dynamic Actions** (transitions and movements): -- Labeled from the **first frame** where the motion begins -- End when the person reaches a **resting state** or begins a new action -- If one motion is followed by another, the transition occurs at the first frame showing movement not explained by the previous action - -**Static States** (stationary postures): -- Begin when person **comes to rest** in that posture -- Continue until the next motion begins -- Example for `sitting`: Does not start when the body touches the chair, but when the body loses its tension and settles into the seated position - -### Label Boundaries - -- **Dynamic → Dynamic**: Transition at first frame of new motion -- **Dynamic → Static**: Static begins when movement stops and body settles -- **Static → Dynamic**: Dynamic begins at first frame of movement - -## Demographic Distribution -Rich demographic and scene metadata enables bias analysis and cross-demographic evaluation. - -![Demographic Overview](figures/demographic_overview.png) - -**Note:** Metadata represents generation prompts. Due to generative model biases, actual visual attributes may deviate, particularly for ethnicity and body type. Age and gender are generally more reliable. - -**Scene Variations:** -- Environments: Indoor/outdoor settings -- Camera angles: 4 elevations × 4 azimuths × 2 distances -- Shot types: Static wide and medium-wide - -## Video Data - -**Videos are NOT included in this repository.** This dataset contains only annotations and metadata. - -### Video Specifications +## License -- **Duration:** 5.0625 seconds per clip -- **Frame count:** 81 frames -- **Frame rate:** 16 fps -- **Format:** MP4 (H.264) -- **Resolution:** Variable (synthetic generation) +The annotations and split definitions in this repository are released under [Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License](https://creativecommons.org/licenses/by-nc-sa/4.0/). -### Accessing Videos +The original video data belongs to their respective owners and should be obtained from the original sources. -Videos will be released at a later point in time. Information about access will be provided here when available. +## Contact -When videos become available, they should be organized with the following structure: -``` -video_root/ -├── fall/ -│ ├── fall_ch_001.mp4 -│ ├── fall_ch_002.mp4 -│ └── ... -├── fallen/ -│ ├── fallen_ch_001.mp4 -│ └── ... -└── ... -``` +For questions about the dataset, please contact [david.schneider@kit.edu]. -The `path` field in the CSV corresponds to the relative path without the `.mp4` extension (e.g., `"fall/fall_ch_001"` → `video_root/fall/fall_ch_001.mp4`). - -## License -[![License: CC BY-NC 4.0](https://img.shields.io/badge/License-CC%20BY--NC%204.0-lightgrey.svg)](https://creativecommons.org/licenses/by-nc/4.0/) +## How we used VGG VIA to densely annotate videos for Omnifall +We provide this video to shortly demonstrate how the annotation process was conducted, increasing transparency. Note, that the CMDFall dataset already provides realtively detailed labels, this is not the case for many of the other datasets. -Annotations and metadata released under [CC BY-NC 4.0](https://creativecommons.org/licenses/by-nc/4.0/). Video data is synthetic and subject to separate terms. +