Domes to Drones: Self-Supervised Active Triangulation for 3D Human Pose Reconstruction

Authors: Aleksis Pirinen, Erik Gärtner, Cristian Sminchisescu

NeurIPS 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental In extensive evaluations on complex multi-people scenes filmed in a Panoptic dome, under multiple viewpoints, we compare our active triangulation agent to strong multi-view baselines, and show that ACTOR produces significantly more accurate 3d pose reconstructions.
Researcher Affiliation Collaboration 1Department of Mathematics, Faculty of Engineering, Lund University 2Google Research {aleksis.pirinen, erik.gartner, cristian.sminchisescu}@math.lth.se
Pseudocode No The paper describes the ACTOR agent and its training process in detail, including mathematical formulations for the reward signal, but it does not provide any structured pseudocode block or algorithm figure.
Open Source Code No The paper does not contain any explicit statement about releasing the source code for the methodology described, nor does it provide a link to a code repository.
Open Datasets Yes We study the active triangulation problem in the CMU Panoptic multi-camera framework [17] since its data consists of real videos of people and allows for reproducible experiments.
Dataset Splits Yes We select 20 scenes (343k images) which are split randomly into training, validation and test sets with 10, 4, and 6 scenes, respectively.
Hardware Specification No The paper mentions the 'Panoptic multi-camera dome' as the data source and a 'Crazyflie drone' for a proof-of-concept experiment, but it does not specify the computing hardware (e.g., GPU models, CPU types, memory) used for training or running the main experiments. It only provides runtime measurements.
Software Dependencies No The paper mentions key software components like 'Open Pose 2d pose estimation system [4]' and 'Adam [18]' (an optimizer), but it does not provide specific version numbers for these or other required software libraries (e.g., Python, PyTorch/TensorFlow, CUDA).
Experiment Setup Yes The policy is trained for 75k episodes with learning rate initially set to 5e-7, then halved after 720k steps and again after 1440k steps. The precision parameters (ma, me) of the von Mises distributions are linearly annealed from (1, 10) to (25, 50) during training...