Aligning Silhouette Topology for Self-Adaptive 3D Human Pose Recovery

Authors: Ramesha Rakesh Mugaludi, Jogendra Nath Kundu, Varun Jampani, Venkatesh Babu R

NeurIPS 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experimental results demonstrate our superiority against prior-arts in self-adapting a source trained model to diverse unlabeled target domains, such as a) in-the-wild datasets, b) low-resolution image domains, and c) adversarially perturbed image domains (via UAP).
Researcher Affiliation Collaboration Mugalodi Rakesh1 Jogendra Nath Kundu1 Varun Jampani2 R. Venkatesh Babu1 1Indian Institute of Science, Bangalore 2Google Research
Pseudocode Yes Algorithm 1: Proposed adaptation procedure.
Open Source Code No The paper provides a webpage link (https://sites.google.com/view/align-topo-human) but does not contain an explicit statement that the source code for the described methodology is released or available there. It is not a direct link to a code repository.
Open Datasets Yes Datasets. Moshed [37] CMU-Mo Cap [32] and H3.6M training-set [19] form our unpaired 3D pose data... We use SURREAL [57] to train our synthetic source model... We use a mixture of Human3.6M [19] and MPII [1] as our Real-domain dataset... We perturb the clean Human3.6M [19] samples... We use low-resolution (LR) variants of 3DPW [58] dataset...
Dataset Splits Yes We evaluate on Human3.6M [19] following Protocol-2 [23].
Hardware Specification Yes A single iteration of the iterative fitting procedure takes nearly 12ms on a Titan-RTX GPU.
Software Dependencies No The paper mentions software components like 'Res Net-50' and 'Adam optimizer' but does not provide specific version numbers for any software dependencies or libraries.
Experiment Setup Yes We use the Adam optimizer [25] with a learning rate 1e 6 and batch size of 16, while setting Max Iteropt to 10 and K=4.