AI Paper Atlas

Research map / 3D vision / Human pose estimation

CapeLLM: Support-Free Category-Agnostic Pose Estimation with Multimodal Large Language Models

Junho Kim, Hyungjin Chung, Byung-Hoon Kim

ICCV 2025 · 3D vision › Human pose estimation · approach: Category-level object pose estimation · 6 citations (OpenAlex)

Read the paperCheck 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
add cumulative reasoning at inference with dynamic-round training (ablation)CapeLLM · MP-100PCK@0.20: 96.05 -> 97.28Improved
add cumulative reasoning at inference with fixed-round training (ablation)CapeLLM · MP-100PCK@0.20: 96.98 -> 94.59Hurt
use fixed-round training pairing every image with all keypoints instead of LocLLM-style training (ablation)CapeLLM · MP-100mPCK: 84.00 -> 91.98Improved
use spatial and relational keypoint descriptions instead of vague ones (ablation)CapeLLM · MP-100mPCK: 85.98 -> 91.98Improved
add keypoint descriptions to the instruction (ablation)CapeLLM · MP-100mPCK: 89.86 -> 91.98Improved
add a list of all keypoint names to the instruction (ablation)CapeLLM · MP-100mPCK: 91.98 -> 90.97Hurt
diversify question expressions in instructions (ablation)CapeLLM · MP-100mPCK: 91.98 -> 90.56Hurt
add a conversation outline to instructions (ablation)CapeLLM · MP-100mPCK: 91.98 -> 90.63Hurt
randomly replace descriptions during training when descriptions are removed at inference (ablation)CapeLLM · MP-100mPCK: 85.69 -> 88.84Improved
use 4 keypoints per round instead of 2 (ablation)CapeLLM · MP-100mPCK: 89.11 -> 91.98Improved
fine-tune with LoRA instead of keeping the model frozen (ablation)CapeLLM · MP-100mPCK: 88.39 -> 91.98Improved
use full-parameter fine-tuning instead of LoRA (ablation)CapeLLM · MP-100mPCK: 91.98 -> 38.59Hurt

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

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