Keypoint Confidence Analysis for RTM-tri and MP-tri Pose Estimation Models

Naoto Ienaga published a dataset on figshare in April 2026 detailing the percentage of low-confidence keypoints in pose estimation. The data, stored in a 5.5 KB XLS file, compares confidence values below 0.3 for keypoints detected by RTM-tri and MP-tri models, with shoulder keypoints present in all frames.

Use Cases

• Benchmarking model confidence between RTM-tri and MP-tri based on the described low-confidence keypoint percentages.
• Identifying systematic failure modes in pose estimation based on the count and distribution of low-confidence keypoints.
• Analyzing the consistency of shoulder keypoint detection across frames as mentioned in the description.

Strengths

• Data is focused on a specific, measurable metric: keypoint confidence below a 0.3 threshold.
• Provides a direct comparison between two named models, RTM-tri and MP-tri.
• File size is 5.5 KB, ensuring quick download and easy inspection.

Limitations

• Row count is unknown, which may limit suitability assessment.
• Column-level documentation is absent; field semantics must be inferred after download.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

figshare, author Naoto Ienaga.

Collection Method

Likely derived from the output of RTM-tri and MP-tri pose estimation models.

Time Range

null

Freshness

Last updated 2026-04-24 17:23:12; freshness should be verified.

Geography

null