Sora AI Video Watermark Detection Frames

164 frame images extracted from Sora AI-generated videos, annotated for watermark detection. The dataset, created by LLinked and last updated in October 2025, is formatted for YOLOv11 object detection models. It is partitioned into 124 training, 21 validation, and 19 test images.

Use Cases

• Train object detection models like YOLOv11 to locate watermark bounding boxes in video frames.
• Benchmark the performance of watermark detection algorithms on the provided 19-image test set.
• Fine-tune pre-trained detectors using the 124 annotated training images for improved accuracy on Sora-generated content.
• Validate model generalization by evaluating on the separate 21-image validation set.

Strengths

• 164 total annotated image samples for model development
• Pre-split into training (124), validation (21), and test (19) sets for standardized evaluation
• Follows a standard YOLOv11 annotation format for easy integration

Limitations

• Small sample size of 164 total images, which may limit model generalization
• Single class annotation ('watermark') restricts use to binary detection tasks
• Dataset scope is limited to frames from Sora-generated videos only

Provenance

Source

LLinked on Hugging Face.

Collection Method

Frames extracted from Sora AI-generated videos and annotated with watermark bounding boxes.

Freshness

Last updated on 2025-10-13.