[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"detail-sidebar-cat-0-en-105":3,"doc-seo-153383-105":59,"doc-detail-153383-en":130},{"code":4,"msg":5,"data":6},0,"success",[7,13,18,23,28,33,38,43,48,51,55],{"id":8,"doc_module":4,"doc_module_name":9,"category_name":10,"show_sort_weight":11,"slug":12},1,"Document","Story & Novel",90,"story-novel",{"id":14,"doc_module":4,"doc_module_name":9,"category_name":15,"show_sort_weight":16,"slug":17},2,"Literature",80,"literature",{"id":19,"doc_module":4,"doc_module_name":9,"category_name":20,"show_sort_weight":21,"slug":22},4,"Exam",70,"exam",{"id":24,"doc_module":4,"doc_module_name":9,"category_name":25,"show_sort_weight":26,"slug":27},5,"Comic",60,"comic",{"id":29,"doc_module":4,"doc_module_name":9,"category_name":30,"show_sort_weight":31,"slug":32},6,"Technology",50,"technology",{"id":34,"doc_module":4,"doc_module_name":9,"category_name":35,"show_sort_weight":36,"slug":37},7,"Healthcare",40,"healthcare",{"id":39,"doc_module":4,"doc_module_name":9,"category_name":40,"show_sort_weight":41,"slug":42},8,"Research & Report",30,"research-report",{"id":44,"doc_module":4,"doc_module_name":9,"category_name":45,"show_sort_weight":46,"slug":47},9,"Religion & Spirituality",20,"religion-spirituality",{"id":46,"doc_module":4,"doc_module_name":9,"category_name":49,"show_sort_weight":46,"slug":50},"World Cup","world-cup",{"id":52,"doc_module":4,"doc_module_name":9,"category_name":53,"show_sort_weight":52,"slug":54},10,"Lifestyle","lifestyle",{"id":56,"doc_module":4,"doc_module_name":9,"category_name":57,"show_sort_weight":24,"slug":58},19,"General","general",{"code":4,"msg":60,"data":61},"ok",{"site_id":62,"language":63,"slug":64,"title":65,"keywords":66,"description":67,"schema_data":68,"social_meta":123,"head_meta":125,"extra_data":127,"updated_unix":129},105,"en","domes-to-drones-self-supervised-active-triangulation-for-3d-human-pose-reconstruction-supplementary-material","Domes to Drones: Self-Supervised Active Triangulation for 3D Human Pose Reconstruction - Supplementary Material","","Supplementary material detailing the ACTOR model and experimental setup for self-supervised 3D human pose reconstruction from 2D pose estimates. It describes the full network architecture, implementation choices in Caffe and MATLAB with OpenPose feature inputs, and the training pipeline including hyperparameter search with Hyperdock and robustness checks across random seeds. It further explains how multiple people are matched over space and time using siamese instance features, Hungarian assignment, and L2 matching costs, including dataset-related evaluation signals.",{"@graph":69,"@context":122},[70,84,105],{"@type":71,"itemListElement":72},"BreadcrumbList",[73,77,79,82],{"item":74,"name":75,"@type":76,"position":8},"https://docshare.wps.com","Home","ListItem",{"item":78,"name":9,"@type":76,"position":14},"https://docshare.wps.com/document/",{"item":80,"name":30,"@type":76,"position":81},"https://docshare.wps.com/document/technology/",3,{"item":83,"name":65,"@type":76,"position":19},"https://docshare.wps.com/document/domes-to-drones-self-supervised-active-triangulation-for-3d-human-pose-reconstruction-supplementary-material/153383/",{"url":83,"name":65,"@type":85,"image":86,"author":91,"headline":65,"publisher":94,"fileFormat":97,"inLanguage":63,"description":67,"dateModified":98,"datePublished":99,"encodingFormat":97,"isAccessibleForFree":100,"interactionStatistic":101},"DigitalDocument",{"url":87,"@type":88,"width":89,"height":90},"https://docshare.wps.com/thumbnails/domes-to-drones-self-supervised-active-triangulation-for-3d-human-pose-reconstruction-supplementary-material/153383.png","ImageObject",300,407,{"name":92,"@type":93},"Skyler","Person",{"url":74,"name":95,"@type":96},"DocShare","Organization","application/pdf","2026-10-06","2026-08-27",true,{"@type":102,"interactionType":103,"userInteractionCount":44},"InteractionCounter",{"@type":104},"ViewAction",{"@type":106,"mainEntity":107},"FAQPage",[108,114,118],{"name":109,"@type":110,"acceptedAnswer":111},"What does the ACTOR supplementary material cover?","Question",{"text":112,"@type":113},"It provides more insight into the ACTOR model and the experimental setup, including network architecture, hyperparameters, multi-person matching, and additional dataset details.","Answer",{"name":115,"@type":110,"acceptedAnswer":116},"How is the ACTOR network implemented and what input does it use?",{"text":117,"@type":113},"The ACTOR model is implemented in Caffe and MATLAB. It uses deep feature maps produced by OpenPose (e.g., conv4_4_CPM) as input to the policy network.",{"name":119,"@type":110,"acceptedAnswer":120},"How are multiple people matched in space and time?",{"text":121,"@type":113},"The method computes instance-sensitive deep features for detections and matches them stably using the Hungarian algorithm with an L2 distance-based matching cost.","https://schema.org",{"og:url":83,"og:type":124,"og:title":65,"og:site_name":95,"og:description":67},"article",{"robots":126,"canonical":83},"index,follow",{"doc_id":128,"site_id":62},153383,1787875136,{"code":4,"msg":5,"data":131},{"doc_id":128,"user_id":132,"nickname":92,"user_avatar":133,"doc_module":4,"category_id":29,"category_name":30,"doc_title":65,"doc_description":67,"doc_content":134,"file_id":135,"file_url":136,"file_type":137,"file_size":138,"view_count":44,"is_deleted":4,"is_public":8,"is_downloadable":8,"audit_status":8,"page_count":19,"language":139,"language_code":63,"site_id":62,"html_lang":63,"table_of_contents":140,"faqs":141,"seo_title":142,"seo_description":67,"update_tm":129,"read_time":52},2336464648746,"https://ap-avatar.wpscdn.com/davatar_276721f389ce27ea32af1340a28f341c","Domes to Drones: Self-Supervised Active Triangulation for 3D Human Pose Reconstruction Supplementary Material  \nAleksis Pirinen 1 􀀃, Erik Gärtner 1􀀃 and Cristian Sminchisescu 1 ;2  \n1Department of Mathematics, Faculty of Engineering, Lund University  \n2 Google Research  \n{aleksis .pirinen, erik .gartner, cristian .sminchisescu}@math .lth .se  \nThis supplementary provides more insight into our ACTOR model and experimental setup. Section §1 describes the details of the network architecture, implementation, and hyperparameters. §2 elaborateson how we match 2d pose estimates in space and time using instance features. In §3 we provide 2dreprojection errors onto 2d OpenPose [2] estimates on the Panoptic test splits. Finally, §4 describes further dataset details.  \n1 Model Architecture  \nSee Fig. 1 for a full description of the ACTOR network architecture. ACTOR was implemented in Caffe [5] and MATLAB. We used an open-source TensorFlow implementation of OpenPose2. All code and pre-trained weights have been made publicly available.3  \nFigure 1: ACTOR policy architecture. A multi-people 2d pose estimation system (here OpenPose, but any similar deep system would would work) processes an input image. The deep feature maps Bt produced by OpenPose (conv4_4_CPM) is fed into the ACTOR policy network and is processed by two convolutional layers with ReLU-activations. The first and second convolutional layers both have 3 􀀂 3 kernels with stride 1 . Their output dimensions are 8 􀀂 39 􀀂 21 and 4 􀀂 18 􀀂 9, respectively. The max pooling layer has a 2 􀀂 2 kernel with stride 2 . The output from the second convolutional layer is then concatenated with agent-centric camera rig information about previously visited cameras relative to current position (Rig), and auxiliary information such as the number of joints triangulated and number of people detected in the view (Aux) . The flattened and concatenated data is then fed into three fully-connected layers with tanh-activations with 1024, 512 and 2 neurons respectively. The final output is scaled by two constants to produce radial angles on the viewing sphere.  \n1.1 Hyperparamters  \nHyperparameter search was performed using two powerful workstations equipped with several NVIDIA Titan V100 GPU:s. Training a single model for 40k episodes took about 32 hours using one GPU and to speed up results while searching for optimal hyperparameters we trained several model configurations in parallell using Hyperdock [3] . The most important parameters for training ACTOR  \n􀀃 Denotes equal contribution, order determined by coin flip.  \n2 [https://gist.github.com/alesolano/b073d8ec9603246f766f9f15d002f4f4](https://gist.github.com/alesolano/b073d8ec9603246f766f9f15d002f4f4)  \n3 [https://github.com/ErikGartner/actor](https://github.com/ErikGartner/actor)  \n33rd Conference on Neural Information Processing Systems (NeurIPS 2019), Vancouver, Canada.  \nwere learning rate, precision of the the von Mises (ma , me ) and the annealing rate of the precision. See Table 1 for a summary of the values tested for these hyperparameters. In total we trained around 10 different versions of the final model with varying hyperparameters and evaluated each of them on the validation set. Finally, the best model was evaluated on the test dataset and retrained with four additional random seeds to measure the model's sensitivity to the random seed (the model is not very sensitive as indicated in Fig. 2, main paper) .  \n\n| Hyperparameter | Attempted values |\n| --- | --- |\n| Learning Rate | {1e-7, 5e-7, 1e-6, 5e-6} |\n| von Mises precision | {(1, 10) ! (25, 50),(10, 50) ! (20, 100),(10,50) ! (100, 500)} |\n\nTable 1: The values tested for the most important hyperparameters when training ACTOR. The finaland best values are highlighted in bold. For the von Mises precisions, the arrow indicates linear annealing performed during training (e.g. from (ma ; me ) = (1 ; 10) to (ma ; me ) = (25 ; 50) for the best configuration) .  \n2 Matching Multiple People  \nACTOR rec","cbCaivkCvKXEkOax","https://ap.wps.com/l/cbCaivkCvKXEkOax","pdf",180394,"English","# Model Architecture\n## Hyperparameters\n# Matching Multiple People","[{\"question\":\"What does the ACTOR supplementary material cover?\",\"answer\":\"It provides more insight into the ACTOR model and the experimental setup, including network architecture, hyperparameters, multi-person matching, and additional dataset details.\"},{\"question\":\"How is the ACTOR network implemented and what input does it use?\",\"answer\":\"The ACTOR model is implemented in Caffe and MATLAB. It uses deep feature maps produced by OpenPose (e.g., conv4_4_CPM) as input to the policy network.\"},{\"question\":\"How are multiple people matched in space and time?\",\"answer\":\"The method computes instance-sensitive deep features for detections and matches them stably using the Hungarian algorithm with an L2 distance-based matching cost.\"}]","Domes to Drones: Self-Supervised Active Triangulation for 3D Human Pose Reconstruction - Supplementary Material | PDF"]