---
license: mit
task_categories:
- video-classification
- other
language:
- en
tags:
- synthetic-data
- video
- circle-behavior
- physics-simulation
- machine-learning
- videomae
- representation-learning
size_categories:
- 100K<n<1M
pretty_name: Circle Behavior SFR 8-Frame Dataset
---

# Circle Behavior SFR 8-Frame Video Dataset

## Dataset Description

This dataset contains 310,000 synthetic videos of circle behavior simulations for machine learning research. Each video shows a circle with specific behavioral patterns: **S**witch color (infinite), **F**ixed size (20px), **R**andom speed.

**Pattern Code**: SFR = **S**witch Color, **F**ixed Size, **R**andom Speed

## Dataset Structure

- **Training videos**: 300,000
- **Test videos**: 10,000 
- **Total videos**: 310,000
- **Frames per video**: 8
- **Resolution**: 256×256 pixels
- **Storage format**: HDF5 with MP4-encoded videos
- **Total size**: ~2.1GB (train: 2.0GB, test: 69MB)

## Sample Visualizations

### Video Sequence Examples

Here are sample 8-frame sequences showing the SFR behavior pattern (horizontal temporal progression):

**Sample 1 - Color Switch Pattern:**

![Sample 1 Sequence](sample_00_sequence.jpg)

**Sample 2 - Different Color Transitions:**

![Sample 2 Sequence](sample_01_sequence.jpg)

**Sample 3 - Varied Movement Patterns:**

![Sample 3 Sequence](sample_02_sequence.jpg)

### Key Visual Features

- **Color Switching**: Colors change unpredictably at different time points
- **Fixed Size**: All circles maintain consistent 20-pixel radius
- **Random Speed**: Movement velocity varies between videos
- **Bounds Safety**: All trajectories stay within frame boundaries

## Data Format

The dataset is stored in HDF5 format with the following structure:
- `train.h5`: Training data (2.0GB)
- `test.h5`: Test data (69MB)

Each video sequence contains 8 frames showing circle behavior with:
- **Color Behavior**: Infinite color switching (unpredictable changes)
- **Size Behavior**: Fixed at 20 pixels
- **Speed Behavior**: Random movement patterns

## Usage

### Python Usage
```python
import h5py
import numpy as np

# Load training data
with h5py.File('train.h5', 'r') as f:
    videos = f['videos'][:]  # Shape: (300000, 8, 256, 256, 3)
    
# Load test data  
with h5py.File('test.h5', 'r') as f:
    test_videos = f['videos'][:]  # Shape: (10000, 8, 256, 256, 3)
```

### Command Line Download
```bash
# Download entire dataset
huggingface-cli download your-username/circle-behavior-sfr-8f-300k \
  --repo-type dataset \
  --local-dir ./circle_behavior_dataset

# Download specific file
huggingface-cli download your-username/circle-behavior-sfr-8f-300k \
  train.h5 \
  --repo-type dataset
```

## Dataset Statistics

| Attribute | Value |
|-----------|-------|
| Videos | 310,000 total |
| Resolution | 256×256 pixels |
| Frames/Video | 8 frames |
| Color Behavior | Switch (infinite) |
| Size Behavior | Fixed (20px) |
| Speed Behavior | Random |
| Format | HDF5 + MP4 |
| Size | 2.1GB |

## Technical Details

- **Color Space**: RGB (0-255)
- **Frame Rate**: Variable (physics-based)
- **Compression**: MP4 encoding within HDF5
- **Validation**: Bounds-safe generation verified
- **Quality**: No corrupted frames, all videos validated