Keypoints as Dynamic Centroids for Unified Human Pose and Segmentation

Authors: Niaz Ahmad, Jawad Khan, Kang G. Shin, Youngmoon Lee, Guanghui Wang

IJCAI 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Our experimental evaluations focus on crowded and occluded cases using the Crowd Pose, OCHuman, and COCO benchmarks, demonstrating KDC s effectiveness and generalizability in challenging scenarios in terms of both accuracy and runtime performance.
Researcher Affiliation Academia Niaz Ahmad1 , Jawad Khan2 , Kang G. Shin3 , Youngmoon Lee4 and Guanghui Wang1 1Department of Computer Science, Toronto Metropolitan University, Canada 2 School of Computing, Gachon University, Republic of Korea, 3Department of Electrical Engineering and Computer Science, University of Michigan, USA 4 Department of Robotics, Hanyang University, Republic of Korea EMAIL, EMAIL, EMAIL, EMAIL
Pseudocode No The paper describes its technical approach in text and mathematical formulas across sections 3.1, 3.2, and 3.3. There are no explicitly labeled pseudocode or algorithm blocks presented in the paper.
Open Source Code Yes Our implementation is available at: https://sites.google.com/view/ niazahmad/projects/kdc.
Open Datasets Yes We evaluate KDC on COCO [Lin et al., 2014], Crowd Pose [Li et al., 2019], and OCHuman [Zhang et al., 2019] benchmarks.
Dataset Splits Yes The model is trained end-to-end using the COCO keypoint and segmentation training set, and ablations are conducted on the COCO val set.
Hardware Specification Yes Models are tested on a single Titan RTX.
Software Dependencies No The paper mentions "Adam optimizer is employed" but does not specify any software libraries or frameworks (e.g., PyTorch, TensorFlow) with version numbers.
Experiment Setup Yes Hyperparameters for training are: learning rate = 0.1 e 4, image size = 401 401, batch size = 4, training epochs = 400, and Adam optimizer is employed. Various transformations are applied during model training, such as scale, flip, and rotate operations. Unless otherwise specified, a disk DR s radius is set to be R = 32.