Datasets:
Tasks:
Video Classification
Formats:
csv
Languages:
English
Size:
10K - 100K
ArXiv:
Tags:
medical
License:
Update README.md
Browse files
README.md
CHANGED
@@ -307,8 +307,8 @@ The repository is organized as follows:
|
|
307 |
Each label file in the `labels/` directory follows this format:
|
308 |
|
309 |
```
|
310 |
-
path,label,start,end,subject,cam
|
311 |
-
path/to/clip,class_id,start_time,end_time,subject_id,camera_id
|
312 |
```
|
313 |
|
314 |
Where:
|
@@ -328,12 +328,14 @@ Where:
|
|
328 |
- `end`: End time of the segment (in seconds)
|
329 |
- `subject`: Subject ID
|
330 |
- `cam`: Camera view ID
|
|
|
331 |
|
332 |
-
For OOPS-Fall, only fall segments and non-fall segments are labeled; non-falls are labels as "other", independent of the underlying content, as long as it is not a fall.
|
|
|
333 |
|
334 |
### Split Format
|
335 |
|
336 |
-
Split files in the `splits/` directory list the video segments included in each partition
|
337 |
|
338 |
```
|
339 |
path
|
@@ -345,7 +347,7 @@ path/to/clip
|
|
345 |
We provide multiple evaluation configurations via the `dataset.yaml` file:
|
346 |
|
347 |
### Basic Configurations
|
348 |
-
- `default`: Access to all dataset labels
|
349 |
- `cs`: Cross-subject splits for all datasets
|
350 |
- `cv`: Cross-view splits for all datasets
|
351 |
|
@@ -359,77 +361,6 @@ We provide multiple evaluation configurations via the `dataset.yaml` file:
|
|
359 |
- `cs-staged-wild`: Train and validate on staged datasets with cross-subject splits, test on OOPS-Fall
|
360 |
- `cv-staged-wild`: Train and validate on staged datasets with cross-view splits, test on OOPS-Fall
|
361 |
|
362 |
-
## Usage
|
363 |
-
|
364 |
-
To use this dataset with the Hugging Face datasets library:
|
365 |
-
|
366 |
-
```python
|
367 |
-
from datasets import load_dataset
|
368 |
-
|
369 |
-
# Load the entire dataset with default configuration
|
370 |
-
dataset = load_dataset("omnifall")
|
371 |
-
|
372 |
-
# Use cross-subject (CS) evaluation protocol
|
373 |
-
cs_dataset = load_dataset("omnifall", "cs")
|
374 |
-
print(f"Train: {len(cs_dataset['train'])} samples")
|
375 |
-
print(f"Validation: {len(cs_dataset['validation'])} samples")
|
376 |
-
print(f"Test: {len(cs_dataset['test'])} samples")
|
377 |
-
|
378 |
-
# Use cross-view (CV) evaluation protocol
|
379 |
-
cv_dataset = load_dataset("omnifall", "cv")
|
380 |
-
|
381 |
-
# Use staged-to-wild evaluation protocol (train on staged datasets, test on OOPS)
|
382 |
-
staged_to_wild = load_dataset("omnifall", "cs-staged-wild")
|
383 |
-
|
384 |
-
# Use individual dataset
|
385 |
-
cmdfall = load_dataset("omnifall", "cmdfall")
|
386 |
-
|
387 |
-
# Access specific fields from the dataset
|
388 |
-
for item in dataset["train"][:5]:
|
389 |
-
print(f"Path: {item['path']}, Label: {item['label']}")
|
390 |
-
```
|
391 |
-
|
392 |
-
## Experiment Examples
|
393 |
-
|
394 |
-
### Cross-Subject Fall Detection
|
395 |
-
|
396 |
-
```python
|
397 |
-
from datasets import load_dataset
|
398 |
-
import torch
|
399 |
-
from torch.utils.data import DataLoader
|
400 |
-
|
401 |
-
# Load the cross-subject evaluation protocol
|
402 |
-
dataset = load_dataset("omnifall", "cs-staged")
|
403 |
-
|
404 |
-
# Preprocess and create dataloaders
|
405 |
-
def preprocess(examples):
|
406 |
-
# Your preprocessing code here
|
407 |
-
return examples
|
408 |
-
|
409 |
-
processed_dataset = dataset.map(preprocess, batched=True)
|
410 |
-
train_dataloader = DataLoader(processed_dataset["train"], batch_size=32, shuffle=True)
|
411 |
-
val_dataloader = DataLoader(processed_dataset["validation"], batch_size=32)
|
412 |
-
test_dataloader = DataLoader(processed_dataset["test"], batch_size=32)
|
413 |
-
|
414 |
-
# Train and evaluate your model
|
415 |
-
```
|
416 |
-
|
417 |
-
### Staged-to-Wild Generalization
|
418 |
-
|
419 |
-
```python
|
420 |
-
from datasets import load_dataset
|
421 |
-
|
422 |
-
# Load the staged-to-wild evaluation protocol
|
423 |
-
dataset = load_dataset("omnifall", "cs-staged-wild")
|
424 |
-
|
425 |
-
# Train on staged data
|
426 |
-
train_data = dataset["train"]
|
427 |
-
val_data = dataset["validation"]
|
428 |
-
|
429 |
-
# Evaluate on wild data
|
430 |
-
wild_test_data = dataset["test"]
|
431 |
-
```
|
432 |
-
|
433 |
## Citation
|
434 |
|
435 |
If you use OmniFall in your research, please cite our paper (will be updated soon):
|
|
|
307 |
Each label file in the `labels/` directory follows this format:
|
308 |
|
309 |
```
|
310 |
+
path,label,start,end,subject,cam,dataset
|
311 |
+
path/to/clip,class_id,start_time,end_time,subject_id,camera_id,dataset_name
|
312 |
```
|
313 |
|
314 |
Where:
|
|
|
328 |
- `end`: End time of the segment (in seconds)
|
329 |
- `subject`: Subject ID
|
330 |
- `cam`: Camera view ID
|
331 |
+
- `dataset`: Name of the dataset
|
332 |
|
333 |
+
For OOPS-Fall, only fall segments and non-fall segments are labeled; non-falls are labels as "other", independent of the underlying content, as long as it is not a fall.
|
334 |
+
Cam and subject ids in OOPS-Fall are -1.
|
335 |
|
336 |
### Split Format
|
337 |
|
338 |
+
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.:
|
339 |
|
340 |
```
|
341 |
path
|
|
|
347 |
We provide multiple evaluation configurations via the `dataset.yaml` file:
|
348 |
|
349 |
### Basic Configurations
|
350 |
+
- `default`: Access to all dataset labels (huggingface loads everything into the `train` split by default.)
|
351 |
- `cs`: Cross-subject splits for all datasets
|
352 |
- `cv`: Cross-view splits for all datasets
|
353 |
|
|
|
361 |
- `cs-staged-wild`: Train and validate on staged datasets with cross-subject splits, test on OOPS-Fall
|
362 |
- `cv-staged-wild`: Train and validate on staged datasets with cross-view splits, test on OOPS-Fall
|
363 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
364 |
## Citation
|
365 |
|
366 |
If you use OmniFall in your research, please cite our paper (will be updated soon):
|