OregonCoastin4K / README.md
Overlaiapp's picture
Update README.md
a48eb91 verified
|
raw
history blame
5.42 kB
---
license: apache-2.0
task_categories:
- text-to-video
size_categories:
- n<1K
configs:
- config_name: default
data_files: data/**/*
language:
- en
tags:
- video
- fine
- finetune
- overlai.ai
- c2pa
- contentcredentials
- opt-in
- 4K
- 4k
- Oregon
- Ocean
- coast
- drone
- aerial
- cinematic
- parallax
---
![OregonCoast4KFineVideo](header.jpg)
# OREGON COAST IN 4K
<!-- Provide a quick summary of the dataset. -->
**"Oregon Coast in 4K"** is a fine tuning text-to-video dataset consisting of dynamic videos captured in 8K resolution on the DJI Inspire 3 and RED Weapon Helium.
## Key Features
- πŸŽ₯ **Oversampled**: Every clip is captured in stunning 8K resolution, delivering rich detail ideal for fine tuning scenic landscapes and ocean dynamics.
- πŸ”„ **Parallax**: Shot using DJI Inspire 3 featuring parallax effects that provide AI models with enhanced context on depth and movement, aiding in spatial awareness training.
- πŸ“Έ **Variance**: Includes close-up details, slow-motion footage of crashing waves, sweeping landscapes, and wildlife shots.
- πŸ“‹ **Detailed Metadata**: Every clip is paired with structured metadata, including creative descriptions, precise camera movements, lens information, field of view calculations, and shot settings, ensuring AI models can fully understand and replicate real-world cinematography with accuracy.
- βš™οΈ **Consistency**: Re-thinking training data at the point of capture by "overshooting" a subject, enabling models to learn more nuanced relationships and views across scenes.
- πŸŒ… **Light**: Shot during early morning and sunset light for optimal color contrast and dynamic range, maximizing visual quality for color and lighting-sensitive tasks.
- πŸ” **Curation**: Curated specifically for machine learning, providing clean, high-quality data for next generation model training.
## Dataset Details
- **Total videos:** 148 Unique videos with 4K and 720p variants. 296 Total.
- **Total size:** 16.06 GB
- **Total duration:** 34 Minutes and 37 Seconds of each resolution variant. 69 minutes total.
- **Video types:** Aerial, Slow Motion, Animals, People, Automotive and Nature
## Technical Details
- **Cameras:** DJI Inspire 3 and RED Weapon Helium
- **Lenses:** DJI DL 24mm and 50mm, Venus Laowa 12mm Cine
- **Resolution:** 3840x2160 and 1280x720 (captured and available in 8192x4320 upon request)
- **Framerate:** 23.976 (captured in 59.94 or 29.97)
- **Codec:** H.264 (captured in ProRes RAW)
- **Data Rate:** 50Mbps (available in higher data rates upon request)
## Metadata
The metadata is made available through a series of parquet files with the following schema:
- `resolution`: Differentiates between either the 4K or 720p version of a video.
- `scene_description`: A natural language description of the scene, detailing the subject, environment, lighting, and mood.
- `shot_type`: The method or perspective of the shot, such as aerial, drone, tripod, or crane.
- `camera_movement`: The type of motion applied to the camera, including dolly, pan, tilt, or other cinematic movements.
- `speed_or_intensity`: The perceived pace or energy of the action within the scene, categorized as slow, medium, or fast.
- `slow_motion_factor`: The percentage of playback speed compared to real-time, indicating how much the footage has been slowed down.
- `current_framerate`: The frame rate of the video file in this dataset, specifying the number of frames displayed per second (e.g., 23.976).
- `current_framerate`: The frame rate at which the footage was recorded, indicating the number of frames captured per second (e.g., 29.97, 59.94).
- `shutter_angle`: The angle of the camera's shutter, controlling motion blur and exposure time relative to the capture framerate (e.g., 180Β° for natural motion blur).
- `camera_model`: The specific make and model of the camera used for capturing the footage (e.g., DJI Inspire 3).
- `sensor_width`: The width of the camera sensor in millimeters, defining the horizontal dimension of the active imaging area (e.g., 33.5mm).
- `sensor_height`: The height of the camera sensor in millimeters, defining the vertical dimension of the active imaging area (e.g., 19mm).
- `lens`: The focal length and type of lens used for capturing the footage, which determines the field of view and depth of field (e.g., 18mm wide-angle lens).
- `horizontal_fov`: The angular field of view across the width of the frame, measured in degrees (e.g., 85.9Β°).
- `vertical_fov`: The angular field of view across the height of the frame, measured in degrees (e.g., 55.6Β°).
- `diagonal_fov`: The angular field of view across the diagonal of the frame, providing the widest perspective measurement (e.g., 93.9Β°).
- `duration`: The total length of the video clip, measured in seconds (e.g., 14.1).
- `size`: The file size of the video, measured in bytes (e.g., 21,870,667 bytes).
### Content Credentials
Each video in the "Oregon Coast in 4K" dataset contains **C2PA Content Credentials** metadata and can be used with the [Content Credentials Space by Truepic](https://huggingface.co/spaces/Truepic/ai-content-credentials) to begin testing model provenance.
## What is Overlai.ai?
[Overlai.ai](https://overlai.ai/) specializes in 8K+ fine tuning datasets with the goal of improving photo & video models. Contact us for the complete dataset list or access to this dataset in full resolution.
## Contact
*hello@overlai.app*