SpeciesNet Analysis of a 2013 Wildlife Video with AI Detection Results

90 full frames and 9 cropped frames from video VID00010 recorded on 2013-07-29 were analyzed using the MegaDetector and SpeciesNet AI systems. The dataset, created by John Kosinski and last updated in April 2026, includes JSON outputs, annotated images, and a structured table of detection outcomes. The analysis documents the AI's response to a subject under conditions of severe motion blur and unfavorable lighting.

Use Cases

• Benchmarking object detector robustness based on performance under motion blur and occlusion described in the analysis
• Studying classifier behavior based on the SpeciesNet 'abstention' outcome for an unknown subject
• Evaluating AI as a corroboration tool based on the comparison of machine detection to human observation
• Analyzing detection confidence trends based on the low-confidence results reported under challenging conditions

Strengths

• Includes 90 full frames and 9 specifically cropped frames for focused analysis
• Provides multiple data formats: JSON outputs, annotated JPG/PNG images, a structured XLSX table, and a PDF summary
• Contains on-site size verification data comparing the subject to a 6-foot human reference

Limitations

• Row count for the structured table is unknown, which may limit suitability assessment
• The analysis is based on a single video (VID00010) from 2013, limiting generalizability
• Column-level documentation for the structured table is absent; field semantics must be inferred after download

Provenance

Source

John Kosinski via figshare

Collection Method

Frame-by-frame analysis of a video using MegaDetector and SpeciesNet AI tools, supplemented by a follow-up site visit for verification.

Time Range

Video recorded on 2013-07-29; analysis published in 2026.

Freshness

Last updated 2026-04-18 00:17:15; freshness should be verified as the underlying video is from 2013.

Geography

null