Sora Video Watermark Detection Frames

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

Use Cases

• Train YOLOv11 models to detect the 'watermark' class in Sora-generated video frames.
• Benchmark object detection performance on a single-class task using the provided 124-image training set.
• Validate watermark detection models on the dedicated set of 21 validation images.
• Test final model accuracy on the held-out set of 19 test images.

Strengths

• Contains 164 annotated frame images specifically from Sora-generated videos.
• Provides predefined splits of 124 training, 21 validation, and 19 test images for structured evaluation.
• Follows a standard YOLOv11 format for ease of use with common object detection pipelines.

Limitations

• Small sample size of 164 total images limits model generalizability.
• Single class ('watermark') restricts use to binary detection tasks.
• Dataset scope is narrow, focusing only on watermarks from one AI video generator (Sora).

Provenance

Source

Frames extracted from Sora AI-generated videos.

Collection Method

Frame extraction and manual annotation for watermarks.

Freshness

Last updated on 2025-10-13.