Detection Moving MNIST Easy: Synthetic Video Sequences for Object Tracking

A synthetic video dataset created by Max-Ploter for object detection and tracking tasks. It features moving MNIST digits with 1-10 digits per sequence, 20 frames per video, and per-frame annotations including digit labels and center coordinates. The dataset was last updated on Hugging Face on 2025-04 09.

Use Cases

• Train object detection models based on per-frame digit labels and bounding box coordinates.
• Benchmark multi-object tracking algorithms based on sequences with 1-10 moving digits.
• Develop models for digit recognition in motion based on grayscale 128x128 video frames.
• Test model robustness to linear trajectories with small random translations.

Strengths

• Provides per-frame annotations including digit labels (0-9) and center coordinates.
• Features controlled sequences with 1-10 digits, 20 frames per video, and consistent 128x128 resolution.
• Digit size is standardized at 28x28 pixels within the frames.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Data is synthetic, which may limit real-world applicability.

Provenance

Source

Hugging Face dataset repository by author Max-Ploter.

Collection Method

Synthetically generated, as described.

Time Range

null

Freshness

Last updated 2025-04-09 08:33:53; freshness should be verified.

Geography

null