AI Paper Atlas

Research map / Detection and segmentation / Open-set and few-shot detection

Few-Shot Keypoint Detection With Uncertainty Learning for Unseen Species

Changsheng Lu, Piotr Koniusz

CVPR 2022 · Detection and segmentation › Open-set and few-shot detection · approach: Few-shot object detection · 36 citations (OpenAlex)

Check your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
Use FSKD with limb-interpolated auxiliary keypoints and multi-scale uncertainty-aware localization instead of the baselineFew-shot keypoint detection · Animal Pose, unseen species, novel keypoints (Avg)PCK@τ=0.1: 22.11 -> 44.75Improved
Use FSKD with limb-interpolated auxiliary keypoints and multi-scale uncertainty-aware localization instead of the baselineFew-shot keypoint detection · CUB, unseen species, novel keypointsPCK@τ=0.1: 66.12 -> 77.89Improved
Use FSKD with limb-interpolated auxiliary keypoints and multi-scale uncertainty-aware localization instead of the baselineFew-shot keypoint detection · NABird, unseen species, novel keypointsPCK@τ=0.1: 39.14 -> 56.04Improved
Use limb interpolation rather than random auxiliary-keypoint pathsFSKD · CUB, unseen species, novel keypointsPCK@τ=0.1: 77.90 vs 77.89No clear effect
Add uncertainty-aware localization to the baseline (ablation)FSKD · Animal Pose, CUB, and NABird, unseen species novel keypointsPCK@τ=0.1: 22.11 -> 24.17; 66.12 -> 68.29; 39.14 -> 41.16Improved
Add auxiliary keypoints for training (ablation)FSKD with UC-GBL · Animal Pose, CUB, and NABird, unseen species novel keypointsPCK@τ=0.1: 24.17 -> 41.70; 68.29 -> 74.50; 41.16 -> 51.62Improved
Add multi-keypoint offset regression loss for grouped keypoints (ablation)FSKD with UC-GBL and auxiliary keypoints · Animal Pose, CUB, and NABird, unseen species novel keypointsPCK@τ=0.1: 41.70 -> 42.60; 74.50 -> 76.25; 51.62 -> 54.27Improved
Use multi-scale UC-GBL rather than single-scale UC-GBL at scale 8 (ablation)FSKD with auxiliary keypoints · Animal Pose, CUB, and NABird, unseen species novel keypointsPCK@τ=0.1: 42.60 -> 44.75; 76.25 -> 77.89; 54.27 -> 56.04Improved
Use HRNet-W48 rather than ResNet50 as the backbone (ablation)FSKD (default) · Animal Pose, unseen species, novel keypoints (Avg)PCK@τ=0.1: 44.75 -> 48.81Improved
Increase support examples from 1-shot to 5-shot (ablation)FSKD (default) · Animal Pose, CUB, and NABird, unseen species novel keypointsPCK@τ=0.1: 10.42%; 2.23%; 7.50%Improved
Increase the number of base keypoint types used for training (ablation)FSKD · CUB and NABird, novel keypointsPCK@τ=0.1Improved
Use Gaussian pooling instead of integer indexing to extract keypoint features (ablation)FSKD · Animal Pose, unseen cat and dog speciesPCK@τ=0.1: 48.11 -> 52.36; 40.60 -> 47.94Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy