{"id":2643,"date":"2022-11-18T16:54:07","date_gmt":"2022-11-18T16:54:07","guid":{"rendered":"https:\/\/salarydistribution.com\/machine-learning\/2022\/11\/18\/see-a-sea-change-3d-researchers-bring-naval-history-to-life\/"},"modified":"2022-11-18T16:54:07","modified_gmt":"2022-11-18T16:54:07","slug":"see-a-sea-change-3d-researchers-bring-naval-history-to-life","status":"publish","type":"post","link":"https:\/\/salarydistribution.com\/machine-learning\/2022\/11\/18\/see-a-sea-change-3d-researchers-bring-naval-history-to-life\/","title":{"rendered":"See a Sea Change: 3D Researchers Bring Naval History to Life"},"content":{"rendered":"
\n

Museumgoers will be able to explore two sunken WWII ships as if they were scuba divers on the ocean floor, thanks to work at Curtin University in Perth, Australia.<\/p>\n

Exhibits in development, for display in Australia and potentially further afield, will use exquisitely detailed 3D models the researchers are creating to tell the story of one of the nation\u2019s greatest naval battles.<\/p>\n

On Nov. 19, 1941, Australia\u2019s HMAS Sydney<\/i> (II) and Germany\u2019s HSK Kormoran <\/i>lobbed hundreds of shells in a duel that lasted less than an hour. More than 700 died, including every sailor on the Sydney.<\/i> Both ships sank 8,000 feet, 130 miles off the coast of Western Australia, not to be discovered for decades.<\/p>\n

<\/p>\n

\"Sydney,<\/p>\n

<\/a>

HMAS Sydney (II) in 1940. (Photo: Allan C. Green from the State Library of Victoria)<\/figcaption><\/figure>\n

Andrew Woods, an expert in stereoscopic 3D visualization and associate professor at Curtin, built an underwater rig with more than a dozen video and still cameras to capture details of the wrecks in 2015.<\/p>\n

Ash Doshi, a computer vision specialist and senior research officer at Curtin, is developing and running software on NVIDIA GPUs that stitches the half-million pictures and 300 hours of video they took into virtual and printed 3D models.<\/p>\n

3D at Battleship Scale<\/b><\/h2>\n

It\u2019s hard, pioneering work in a process called photogrammetry. Commercially available software maxes out at around 10,000 images.<\/p>\n

\u201cIt\u2019s highly computationally intensive \u2014 when you double the number of images, you quadruple the compute requirements,\u201d said Woods, who manages the Curtin HIVE<\/a>, a lab with four advanced visualization systems.<\/p>\n

\u201cIt would\u2019ve taken a thousand years to process with our existing systems, even though they are fairly fast,\u201d he said.<\/p>\n

When completed next year, the work will have taken less than three years, thanks to systems at the nearby Pawsey Supercomputing Centre<\/a> using NVIDIA V100 and prior-generation GPUs.<\/p>\n

Speed Enables Iteration<\/b><\/h2>\n

Accelerated computing<\/a> is critical because the work is iterative. Images must be processed, manipulated and then reprocessed.<\/p>\n

For example, Woods said a first pass on a batch of 400 images would take 10 hours on his laptop. By contrast, he could run a first pass in 10 minutes on his system with two NVIDIA RTX A6000 GPUs<\/a> awarded through NVIDIA\u2019s Applied Research Accelerator Program<\/a>.<\/p>\n

It would take a month to process 8,000 images on the lab\u2019s fast PCs, work the supercomputer could handle in a day. \u201cRarely would anyone in industry wait a month to process a dataset,\u201d said Woods.<\/p>\n

From Films to VR<\/b><\/h2>\n

Local curators can\u2019t wait to get the Sydney<\/i> and Kormoran<\/i> models on display. Half the comments on their Tripadvisor page already celebrate 3D films the team took of the wrecks.<\/p>\n

The digital models will more deeply engage museumgoers with interactive virtual and augmented reality exhibits and large-scale 3D prints.<\/p>\n

\u201cThese 3D models really help us unravel the story, so people can appreciate the history,\u201d Woods said.<\/p>\n

<\/p>\n

\"Kormoran,<\/p>\n

<\/a>

In a video call, Woods and Doshi show how forces embedded an anchor in the Kormoran\u2019s hull as it sank.<\/figcaption><\/figure>\n

The exhibits are expected to tour museums in Perth and Sydney, and potentially cities in Germany and the U.K., where the ships were built.<\/p>\n

When the project is complete, the researchers aim to make their code available so others can turn historic artifacts on the seabed into rare museum pieces. Woods expects the software could also find commercial uses monitoring undersea pipelines, oil and gas rigs and more.<\/p>\n

A Real-Time Tool<\/b><\/h2>\n

On the horizon, the researchers want to try Instant NeRF<\/a>, an inverse rendering tool NVIDIA researchers developed to turn 2D images into 3D models in real time.<\/p>\n

Woods imagines using it on future shipwreck surveys, possibly running on an NVIDIA DGX System<\/a> on the survey vessel. It could provide previews in near real time based on images gathered by remotely operated underwater vehicles on the ocean floor, letting the team know when it has enough data to take back for processing on a supercomputer.<\/p>\n

\u201cWe really don\u2019t want to return to base to find we\u2019ve missed a spot,\u201d said Woods.<\/p>\n

Woods\u2019 passion for 3D has its roots in the sea.<\/p>\n

\u201cI saw the movie Jaws 3D<\/i> when I was a teenager, and the images of sharks exploding out of the screen are in part responsible for taking me down this path,\u201d he said.<\/p>\n

The researchers released the video below to commemorate the 81st anniversary of the sinking of the WWII ships.<\/p>\n

https:\/\/hive.curtin.edu.au\/SK81st<\/a><\/p>\n<\/div>\n","protected":false},"excerpt":{"rendered":"

https:\/\/blogs.nvidia.com\/blog\/2022\/11\/18\/3d-shipwrecks-perth\/<\/p>\n","protected":false},"author":0,"featured_media":2644,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":[],"categories":[3],"tags":[],"_links":{"self":[{"href":"https:\/\/salarydistribution.com\/machine-learning\/wp-json\/wp\/v2\/posts\/2643"}],"collection":[{"href":"https:\/\/salarydistribution.com\/machine-learning\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/salarydistribution.com\/machine-learning\/wp-json\/wp\/v2\/types\/post"}],"replies":[{"embeddable":true,"href":"https:\/\/salarydistribution.com\/machine-learning\/wp-json\/wp\/v2\/comments?post=2643"}],"version-history":[{"count":0,"href":"https:\/\/salarydistribution.com\/machine-learning\/wp-json\/wp\/v2\/posts\/2643\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/salarydistribution.com\/machine-learning\/wp-json\/wp\/v2\/media\/2644"}],"wp:attachment":[{"href":"https:\/\/salarydistribution.com\/machine-learning\/wp-json\/wp\/v2\/media?parent=2643"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/salarydistribution.com\/machine-learning\/wp-json\/wp\/v2\/categories?post=2643"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/salarydistribution.com\/machine-learning\/wp-json\/wp\/v2\/tags?post=2643"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}