AI Detection Results for 8-Foot-Tall Subject in Wildlife Video

John Kosinski's 2026 analysis documents the frame-by-frame AI detection results for video VID00002 recorded on 2013-07 29. The dataset includes 111 full frames, 111 cropped frames, annotated images, JSON outputs, and a structured table of outcomes. It evaluates the performance of the MegaDetector and SpeciesNet classifiers on a specific subject.

Use Cases

• Benchmarking object detector performance under motion blur and occlusion using the full-frame and cropped-frame image sets.
• Analyzing classifier label alternation between 'animal' and 'human' for the subject in the structured table of detection outcomes.
• Studying cases where a classifier abstains from labeling, as indicated in the SpeciesNet JSON outputs for an unknown species.
• Correlating AI detection bounding boxes from annotated images with on-site height verification data showing an 8-foot-tall subject.

Strengths

• Provides 222 total image frames (111 full, 111 cropped) for comparative analysis.
• Includes multiple data views: raw images, annotated outputs, JSON files, and a structured table.
• Contains concrete on-site verification data stating the subject is approximately 8 feet tall.

Limitations

• Small scale at 53.0 MB, focused on a single video (VID00002) from one date in 2013.
• Analysis is specific to one subject, limiting generalizability for model performance.
• Temporal coverage is limited to a single recording event.

Provenance

Source

John Kosinski via figshare.

Collection Method

Frame-by-frame analysis of a video using MegaDetector and SpeciesNet AI models, supplemented by on-site verification.

Time Range

Primary video recorded on 2013-07-29; analysis updated 2026-04-04.

Freshness

Analysis was last updated on 2026-04-04, but the source video is from 2013-07-29.

Geography

null