LiNDA ZOO is LiNDA's digital animal team with over 20 photorealistic digital animals. Such a company has taken on the challenge of a project to reproduce “Shiba Inu Maru”, which has about 2.5 million Instagram followers, with BULLET RENDER FARM (hereafter BRF) and AMD Radeon ProRender (hereafter RPR).
In this article, I will introduce the project issues and their solutions in two parts. In No. 1, we will introduce the process from 3D scanning of "Shiba Inu Maru" to 3D model creation, rigging, simulation, animation, fur settings, and the creation of the image of "Digimaru".
*This article is a revised version of "DigiMARU: Recreating 'Shiba Inu Maru' at BULLET RENDER FARM" published in the monthly ``CGWORLD + digital video'' vol. 262 (June 2020 issue). .
TEXT_Tomoaki Sawada (Colossus R Studio) EDIT_Miyuki Ogata / Miyuki Ogata (CGWORLD) PHOTO_Mitsuru Hirota / Mitsuru Hirota
▲[Case Study] Digimaru: Fur Rendering
▲From left to right in the front row, production manager Kei Oshiro, CG artist Koyo Oyo, CG artist Yoji Arai, animation director Shinsaku Hashimoto, and CG artist Mirai Asakura. From the left in the back row, Assistant Producer Junya Kimura (LiNDA ZOO), Software Development Engineer Atsushi Yoshimura (AMD Japan), Director Kiyonobu Kitada (LiNDA ZOO), Executive Vice President Michitaka Tokeiji (A.L.I. Technologies) ), Producer Daisuke Kuwabara, System Engineer Tomoya Nakagawa (LiNDA ZOO)
Cloud Rendering Increasingly Necessary with Support for Telework
Render farms are indispensable systems for producing state-of-the-art, realistic, and high-quality CG images. A long time ago, it was common for each CG production to have its own render farm. A certain amount of space was secured to accommodate a large number of computers, a dedicated air conditioner was installed to cool the exhaust heat from the computers, and dedicated engineers were assigned to operate the constantly updated software and systems.
Even if you spend that much money, the main parts that make up a computer evolve every year, so it is inevitable that they will become obsolete as soon as they are in operation. Above all, a considerable amount of electricity is required to maintain the performance of the render farm, but since each office or building has an upper limit on the amount of electricity that can be used, it is a headache for CG production.
The cloud computing render farm is a means to quickly solve the aforementioned problems (initial investment cost, system construction cost, operation management cost, power consumption). However, when it comes to CG video production, existing software is not provided on the premise of cloud computing, and it is only an afterthought. Due to the problems of depreciation of hardware assets and license costs, the current situation is that the switch to render farms using cloud computing is not progressing.
Nevertheless, moving IT infrastructure to the cloud remains an effective means of controlling costs, optimizing resources, and enhancing responsiveness to capture opportunities. Digital transformation using the cloud is also required in response to the new coronavirus infectious disease (COVID-19), including telework. The knowledge gained from this project will help us to do so.
Reproduce "Shiba Inu Maru" instead of an unspecified Shiba Inu
BRF operated by A.L.I. It is a cloud rendering service that performs rendering processing. Beyond the framework of a render farm with a large number of GPUs installed, we are also considering providing a service that matches the demand for computer computing power of various scales required by companies and individuals with idling computers around the world. development is underway.
In other words, BRF aims to evolve into grid computing that distributes and executes processing that requires a large amount of computing power to surplus computers. As part of the promotion of the BRF, we planned a challenge to process animal fur, which has been considered to have a particularly high rendering load, with BRF at high speed.
After that, we entered the stage of considering ``what kind of digital animal to make specifically?'', and A.L.I. It was "Shiba Inu Maru". In this way, the "DigiMARU" project, which reproduces "Shiba Inu Maru" in 3DCG and also produces promotional videos for BRF, has started. This time, the aim was not to reproduce an unspecified Shiba Inu in 3DCG, but to resemble a specific individual named "Shiba Inu Maru", and it can be said that it was a challenging project in this respect as well.
Moreover, the Shiba Inu is a familiar animal to the Japanese people, so if there is something strange about it, they will immediately feel uncomfortable. Artist Mirai Asakura recalled that it was a difficult project in that respect as well.
Photogrammetric 3D scan of Shiba Inu Maru
▲ Shiba Inu Maru is a male dog born in 2007 weighing 18 kg and 52 cm long
▲For 3D scanning, iris from steam-studio is used. This system consisted of 164 Nikon single-lens reflex cameras, and the shutters of all cameras were synchronized with strobes with a flash speed of 1/350. In addition, the scale and color chart were also shot at the same time, and the actual size scale adjustment and color calibration were performed in post-processing. The original application of this system is 3D scanning of the human body, and the upper body can acquire particularly high-definition data. This time, the box horses were piled up to a height that would not frighten the Shiba Inu Maru, ensuring a high resolution when shooting. Despite the concerns of humans, Shiba Inu Maru was quietly riding on top of the horse and being photographed.
▲3D scan of Shiba Inu Maru by photogrammetry
▲The Shiba Inu's fluffy body is reproduced in the RAW scan data. This data was delivered to LiNDA ZOO as 2 types of shape data, 10 million polygons and 3 million polygons, and 8K texture data.
© Shibainu maru
Next page: Creating a 2-layer 3D model from 3D scan data
[[SplitPage]]Creating a 2-layer 3D model from 3D scan data
▲Polygon data obtained by 3D scanning when reproducing an animal with fur in 3DCG cannot be used as is. In order to reproduce the fur with another function (Ornatrix in this project), it was necessary to create a 3D model without the fur. Since the depth from the tip of the fur to the skin cannot be determined from the 3D scan data, they actually touched the body of "Shiba Inu Maru" during the 3D scan to measure the depth of the fur. He also used a photo of Shiba Inu Maru in a shower with fur clinging to his body
▲In addition, the previous 3D model has a two-layer structure, and the inner model is inside. The rig for animation is set to the inner model, and the shaking of the fat layer between the outer and inner models is simulated using Ziva VFX
▲ Only the outside is shown in wireframe. The fat layer model used for calculation during simulation is created by performing Boolean operations on the outer and inner models. The details of the simulation will be described later
Reproduce Shiba Inu Maru's throat shaking and walking style
As mentioned above, Ziva VFX is used to simulate the shaking of the fat layer. The software can create muscles from the skeleton and simulate not only the fat layer but also the epidermis. It is eye-catching that even the swaying of Shiba Inu Maru's tapped throat has been faithfully reproduced.
All the animations in this project were created by hand, and a video of Maru the Shiba Inu taking a walk was used as a reference. The artists at LiNDA ZOO are familiar with animal animation, so the walking style of Maru, the Shiba Inu, has been recreated.
Rig with mGear and fat simulation with Ziva VFX
▲Additional setup after building mGear is automated by a script, one click from the [above] guide You can create the [bottom] rig. If you neglect this automation, it will become an asset that is difficult to maintain, so be careful.
▲The blue part is the fat layer used for simulation calculations. Since the results change greatly depending on the thickness, we repeat the simulation several times to find the optimal thickness.
▲Displaying the simulation resolution of Ziva VFX. The finer the grid, the higher the accuracy, but the longer the calculation time
© Shibainu maru
Next page: Recreating trots, walking, and fur
[[SplitPage]]Reproduction of trot, walking, and fur
▲[Left] Trot reference video/[Right] Walking reference video
▲A comparison of trotting and walking animations. Pink is a trot, blue is a walk
▲ Comparison video of walking reference and animation
▲ Ornatrix work screen. We groomed several parts, such as the head, trunk, and tail. The detailed fur flow is arranged by editing the grooming guide one by one using the selection tool. LiNDA ZOO has used Shave and a Haircut and Yeti to express fur, but this time I used the Maya version of Ornatrix.
▲Adjusting the volume of the fur while comparing it with the 3D scan data
▲Completed "Digimaru". At this stage, rendering is done using Redshift and LiNDA ZOO's in-house render farm
That's it for "Reproducing 'Shiba Inu Maru' with BULLET RENDER FARM No.1 Digimaru/DigiMARU Birth Edition". In "Reproducing "Shiba Inu Maru" with BULLET RENDER FARM No.2 Cloud Rendering", we explain the issues when switching renderers from Redshift to AMD Radeon ProRender, their solutions, and cloud rendering with BULLET RENDER FARM. Matching please see.
© Shibainu maru