THE ROOT ELEMENT OF HUMAN POSES IS RADIAN: MCPRL IS ALL YOU NEED

       

Based on merging H36M and MPII datasets, we construct a new Manifold Cadre Poses withRadian List-systematically (MCPRL) dataset where missing outdoor scenarios, particularly dynamic collision actions are supplemented.

Figure1: Comparison of the MPJPE accuracy of different baselines on unpruned, radian pruned and MCPRL datasets.

Figure2: Comparison of the storage and diversity of unpruned and MCPRL datasets.

Figure3: Schematic diagram of Generalized Radian Pruning.

Figure4: The action distribution of our MCPRL dataset.

Figure5: Examples of our proposed MCPRL Dataset, GRP in blue square represents the rough percentage of original dataset we streamlined.

Figure6: Examples of our annotated CAD Dataset part. Basically all the poses included are during collision or injury.

Figure7: The user interface for collecting 2D pose projection annotations.

Figure8: Rendering interfaces of CAD dataset in Blender.

Reference:
        [1] Ruan, X., & Cheng, Z. (2024 ICME). THE ROOT ELEMENT OF HUMAN POSES IS RADIAN: MCPRL IS ALL YOU NEED.

Download MCPRL Dataset