2. Dataset Specifications

2.1. Generation Criteria

• Maximum video duration: 10 seconds maximum
• Video themes:
  • Walking
  • Exercising
  • Writing
  • Shopping
  • Sleeping
  • Meditating
  • Working
  • Studying
  • Driving
  • Washing
  • Gardening
  • Calling
  • Listening
  • Organizing
  • Planning
  • Relaxing
  • Teaching
• Video size: 512×512 pixels

2.2. Dataset Organization

The dataset follows this structure:

└─ VidData
└─ data
    └─ train
        └─ VidData.csv
└─ video
    └─ ---_iRTHryQ_13_0to241.mp4
    └─ ---agFLYkbY_7_0to303.mp4
    └─ --0ETtekpw0_2_18to486.mp4
└─ readme.md
        

• data/train/: Contains CSV files with metadata associated with the videos.
• video/: Contains the video files.

3. Data Structure

The dataset is stored as a CSV file and includes the following columns:

4. Libraries Used

4.1. Library Examples

Here are some example libraries that can be used when analyzing this data:

• OpenCV: Video manipulation and processing (reading, writing, frame extraction, contour detection, filtering, etc.).
• Scikit-Image: Calculating the Structural Similarity Index (SSIM) for image quality evaluation and various image transformations (segmentation, filtering, etc.).
• NumPy: Efficient manipulation of matrices and arrays, essential for calculations on images and videos.
• Pandas: Managing and structuring metadata associated with videos (e.g., file names, timestamps, annotations).
• Matplotlib/Seaborn: Visualizing analysis results as graphs.

4.2. Installing Dependencies

Follow the instructions below to install the required libraries:

1. Create a requirements.txt file and add the following:

   opencv-python==4.8.1.78 # Video manipulation and processing
   scikit-image==0.22.0 # SSIM calculation and image transformations
   numpy==1.26.2 # Efficient manipulation of matrices and arrays
   pandas==2.1.4 # Managing and structuring metadata
   matplotlib==3.8.2 # Visualizing analysis results
   seaborn==0.12.2 # Advanced visualization with enhanced graphics
                   

2. Run the command: pip install -r requirements.txt

Note: Only include the libraries you need in requirements.txt.

5. Using the Dataset

5.1. Primary Applications

5.1.1. Text-to-Video Generation

• Train models to generate video based on textual input.
• Benchmark performance by comparing generated video against dataset entities.

5.1.2. Video Description Models

• Evaluate models designed to generate textual descriptions from videos.

5.1.3. Temporal Consistency Analysis

• Test model for maintaining smoothness and coherence in video generation.

5.2. Example Workflow

Load the dataset using Python:

import pandas as pd

dataset = pd.read_csv('VidData.csv')
print(dataset.head())

# Access video metadata:
video = dataset.iloc[0]  # First entry
print(f"Video Name: {video['video_name']}")
print(f"Caption: {video['Caption']}")
print(f"Duration: {video['duration_seconds']} seconds")

# Filter video based on motion:
high_motion_videos = dataset[dataset['motion_score'] > 1.0]
print(high_motion_videos)
        

6. File Format

The dataset is delivered in CSV format, with each column representing a video and its metadata.

7. Sample Entry