Preferred Networks Corporate Factbook

Transcript

1. Preferred Networks Corporate Factbook July 2025
2. 2 Preferred Networks (PFN) - Overview Established Headquarters Co-Founders Employees Business Subsidiaries Investors (alphabetical order) March 2014 Tokyo, Japan Toru Nishikawa, Chief Executive Officer Daisuke Okanohara, Chief Technology Officer Approx. 350 Research, development and sales of solutions and products based on AI-related technologies including AI chips, computing infrastructure and generative AI foundation models Matlantis Corporation (Formely “Preferred Computational Chemistry, Inc.”), Preferred Robotics, Inc., Preferred Computing Infrastructure, Inc. Chugai Pharmaceutical, Development Bank of Japan, ENEOS Innovation Partners, Fanuc, Hakuhodo DY Holdings, Hitachi, Mitsubishi Corporation, Mitsubishi UFJ Trust and Banking, Kodansha, Mitsui & Co., Mizuho Bank, NTT, SBI Group, Sekisui House Investment Limited Partnership, Shin-Etsu Chemical Co., Ltd., Sumitomo Mitsui Trust Bank, TBS Innovation Partners Fund III, TEL Venture Capital, Toei Animation, Toyota Motor, Wacom, etc. Mission: Make the real world computable https://www.preferred.jp
3. 3 Research, development and sales of software and hardware for AI Founded in March 2014 PFN Group Companies Programming schools for children Founded in December 2020 Joint venture with Yaruki Switch Group Development and sales of autonomous mobile robots Founded in November 2021 Investors include Amano, SMBC, Asahi Kasei Homes Sales of Matlantis™ universal atomistic simulator Founded in June 2021 Joint venture with ENEOS, invested by Mitsubishi Corporation Provision and operation of AI cloud service Founded in January 2025 Joint venture with Mitsubishi Corporation and IIJ Preferred Computing Infrastructure
4. 4 PFN: Vertically Integrating AI Value Chain Solutions & products Computing infrastructure AI chips PFN combines advanced software and hardware technologies in a vertically integrated approach, covering the entire AI value chain from chips to solution and products. Generative AI foundation models Solutions and products for industries and consumers MN-Core™ MN-Core™ 2 GPU cluster MN-3 (MN-Core™ cluster) PLaMo™ Prime (large language model) PLaMo™ Lite（small language model for edge devices） Cloud-based computing service powered by MN-Core™ 2 Model for simulating material energy PFP Next-gen MN-Core MN-Core™ L1000 for LLM inference (Launch planned in 2027)
5. 5 PFN: Horizontal Application of AI Technologies Generative AI foundation models Society Consumers Augment human capabilities, creative expressions and entertainment experiences Ensure safe and trouble-free life, quality education and healthcare Improve productivity and quality, reduce dependence on individuals, remedy labor shortage Computing infrastructure Industries AI chips PFN applies its vertically-integrated AI technologies horizontally as solutions and products for multiple industries, consumers and society. Plants/ Manufacturing Entertainmen t Robots Retail Pharma- ceuticals/ healthcare Materials/ Chemicals Education Finance
6. 6 PFN Co-Founders Daisuke Okanohara Co-Founder Chief Technology Officer Toru Nishikawa Co-Founder Chief Executive Officer Toru Nishikawa is a co-founder and Chief Executive Officer of Preferred Networks, Inc. (PFN). Prior to founding PFN in 2014 to focus on deep learning, he was the CEO of Preferred Infrastructure, PFN’s predecessor that he co-founded with Daisuke Okanohara and others for developing natural language processing software. Nishikawa obtained his Master’s degree from the University of Tokyo’s Graduate School of Information Science and Technology in 2007. An avid programmer since childhood, he reached the 19th place at the 30th International Collegiate Programming Contest (ACM-ICPC) in 2006. Daisuke Okanohara is a co-founder and the Chief Technology Officer of Preferred Networks, Inc. (PFN). He currently leads PFN’s research projects on foundation models and other AI technologies. Okanohara also serves as CEO of Preferred Computational Chemistry and Preferred Elements. Okanohara obtained his Ph.D degree in Computer Science from the University of Tokyo in 2010. He is also a co-founder of PFN’s predecessor Preferred Infrastructure. He is the first recipient of Gendai no Meiko (contemporary master craftsman) award from Japan’s Ministry of Health, Labour and Welfare as a data scientist.
7. 7 PFN: Path to Founding Left: Daisuke Okanohara Co-Founder Chief Executive Researcher Right: Toru Nishikawa Co-Founder Chief Executive Officer Toru Nishikawa and Daisuke Okanohara meet at University of Tokyo as classmates Japanese government-funded MITOU Program for next-generation software engineers selects Nishikawa’s project Highly Abstract Hardware Description Language and also certifies Okanohara as one of its “supercreators” Nishikawa’s team ranks 19th in the 30th International Collegiate Programming Contest 2005 2001 Nishikawa receives Software Japan Award from Information Processing Society of Japan 2006 2007 Nishikawa receives Master’s degree from Graduate School of Information Science and Technology, The University of Tokyo; Okanohara receives multiple awards including Grand Prix at Young Researcher Association for NLP Studies (YANS), University of Tokyo President’s Award, Outstanding Research from Association for Natural Language Processing 2010 Okanohara receives PhD from the Graduate School of Information Science and Technology, The University of Tokyo 2013 2014 Preferred Infrastructure (PFI) founded (March) PFN founded (March)
8. 8 Motivation- Driven Learn or Die Proud, But Humble Boldly do what no one has done before Code of Conduct: PFN Values PFN members’ code of conduct consists of four principles called PFN Values which defines who they are and what they value in their day-to-day work. Details: https://www.preferred.jp/en/company/values/
9. 9 PFN’s Strengths AI Technology Computing Resources Domain Knowledge • Papers accepted to world’s top AI and machine learning conferences including NeurIPS, AAAI, ICRA • Many Kaggle* Experts or above, contestants at ICPC world championship • First developed deep learning framework Chainer™ in 2015, currently one of top PyTorch contributors • Operates multiple in-house supercomputers • Co-developed proprietary deep learning processor MN-Core™ with Kobe University • In-house supercomputer MN-3 topped Green500* ranking 3 times as world’s most energy-efficient • Joint research experiences with the world’s top companies in automotive, industrial robots, pharmaceuticals, energy, etc. • “Learn or Die” as code of conduct: Aggressive learners of partners’ industry domain knowledge • Many in-house specialists in multiple domains *Kaggle: Machine learning and data science community known for company or government-sponsored competitions to solve data science challenges *ICPC: International Collegiate Programming Contest *Green500: A list of the world’s 500 most powerful supercomputers ranked by energy efficiency (performance per watt)
10. 10 PFN’s AI Technology Credentials Competition Rankings and Awards Machine Learning-Related Awards • #10 out of 1,915 teams: Kaggle RSNA 2024 Lumbar Spine Degenerative Classification - Oct. 2024 • #5 and #9 out of 61 teams: Kaggle Fast or Slow? Predict AI Model Runtime - Nov. 2023 • #4 and #5 (#1 and #2 in Japan) out of 2,662 teams: Kaggle LLM Science Exam - Oct. 2023 • #3 out of 954 teams: Kaggle Google Research - Identify Contrails to Reduce Global Warming - Aug. 2023 • #2 out of 1,231 teams: Kaggle Stable Diffusion - Image to Prompts - May 2023 • #2 out of 936 teams: Kaggle G2Net Detecting Continuous Gravitational Waves - Jan. 2023 • #1 out of 1,220 teams: Kaggle Open Problems - Multimodal Single-Cell Integration - Nov. 2022 • #1 out of 1,588 teams: Kaggle Happywhale competition for whale and dolphin identification - Apr. 2022 • #3 out of 1,547 teams: Kaggle RANZCL CLiP competition for accurate evaluation of catheter placements on chest X-rays - Dec. 2020 • #4 out of 935 teams: Kaggle competition Lyft Motion Prediction for Autonomous Vehicles - Dec. 2020 • #3 out of 193 teams: Kaggle competition Open Images 2019 - Instance Segmentation track - Oct. 2019 • Editor’s Highlights on Nature Communications in two categories (AI and Machine Learning and Materials Science and Chemistry) for paper on core Matlantis™ - May 2022 • Best Paper Award at Human-computer interaction (CHI 2020) - Oct. 2019 • Best Paper Award, Human-Robot Interaction track, IEEE International Conference on Robotics and Automation (ICRA 2018) - May 2018 • Papers accepted to top conferences including NeurIPS, AAAI, ICML, CVPR, ICLR, MIRU Papers at Top Academic Conferences/Journals Details: https://www.preferred.jp/en/company/awards/ Details: https://tech.preferred.jp/en/publications/ • #6 out of 1,499 teams: Kaggle competition RSNA Pneumonia Detection Challenge - Nov. 2018 • #2 out of 454 teams: Kaggle competition Google AI Open Images - Object Detection Track - Sep. 2018
11. 11 • Semi-Grand Prix, Advanced Technology Category, CEATEC Award 2023 for MN-Core™ series • Nihon Keizai Shimbun Award, Nikkei Superior Products and Services Awards for PFN 3D Scan • Editor’s Highlight on Nature Communications for research paper on Matlantis™ • Grand Prix, 18th Japan e-Learning Awards for Playgram™ • #1 (3rd time), Green500 list of world’s most energy-efficient supercomputers for MN-3 • Prime Minister’s Award, 5th Japan Venture Awards • Nihon Keizai Shimbun Award, Nikkei Superior Products and Services Awards for Chainer • Semi-Grand Prix, Industries/Markets Category, CEATEC Award • Minister of Economy, Trade and Industry (METI)'s Award, 3rd Nippon Venture Awards • Technology Award, FT ArcelorMittal Boldness in Business Awards • JEITA Venture Award • Forbes JAPAN CEO of the Year 2016 • #2 for pick task and #4 for stow task in Amazon Picking Challenge PFN: Awards and Recognition 2023 2022 2021 2019 2018 2017 2016
12. 12 PFN’s Organizational Structure Corporate Planning Solutions Products & Services AI Computing Corporate Services Administrative and information security sections Design and operation of computing infrastructure, development and manufacturing of semiconductors Development, sales and maintenance of products for customers, service operations Joint research and development with corporations, creation of new business seeds Materials & Drugs Materials discovery, drug discovery, healthcare Retail Development and operation of products for retail industry Research and development of LLMs and foundation models Shareholders’ Meeting Representative Directors Board of Directors Management Meeting Audit and Supervisory Committee Internal Audit Office (Wholly-owned subsidiary)
13. 13 PFN’s Organizational Structure Corporate Planning Solutions Products & Services AI Computing Corporate Services Administrative and information security sections Design and operation of computing infrastructure, development and manufacturing of semiconductors Development, sales and maintenance of products for customers, service operations Joint research and development with corporations, creation of new business seeds Materials & Drugs Materials discovery, drug discovery, healthcare Retail Development and operation of products for retail industry Research and development of LLMs and foundation models Shareholders’ Meeting Representative Directors Board of Directors Management Meeting Audit and Supervisory Committee Internal Audit Office (Wholly-owned subsidiary)
14. 14 AI Products & Solutions Division LLM Development Division AI Computing Division Materials & Drug Discovery Division Retail Solutions Division Shareholders’ Meeting Representative Directors Board of Directors Management Meeting Audit and Supervisory Committee Internal Audit Office Corporate Planning Division Technology Planning Division Corporate Service Division PFN’s Organizational Structure
15. Preferred Networks Business Outline
16. 16 MN-Core™ Series Roadmap Anticipating the demand growth of semiconductors for AI computing resources, PFN started developing the first generation of AI processor in the MN-Core™ series in 2016. Currently, PFN is developing and producing the series in flagship and generative AI pipelines. Details: https://projects.preferred.jp/mn-core/en/ Flagship AI training, inference and high-performance computing (HPC) For Generative AI LLM inference, etc. MN-Core (TSMC 12nm) Development: 2016- Internal use: 2020- Computing power for external parties: 2023- MN-Core 2 (TSMC 7nm) Test operation: 2023- External sales of servers/ computing power via PFCP™: 2024- Ultra-high-efficiency AI accelerator (Samsung 2nm) R&D under NEDO project from 2024, results to be incorporated in product 2016 2020 2023 2027 Next- Generation Basic agreement with Rapidus and Sakura Internet for Japan-made AI infrastructure MN-Core L1000 Development: 2024- Planned launch: 2027 MN-Core L2000 In development In discussion MN-Core L3000 AI Chips In discussion In discussion
17. 17 MN-Core™ Series: Design Philosophy The MN-Core architecture maximizes the proportion of arithmetic units on the hardware by transferring functions normally allocated to the hardware side to the software side, realizing high performance and energy efficiency. AI Chips Optimization Code generation General-purpose processor MN-Core series Software Register Arith- metic units Command scheduler Cache controller Network control circuit Hardware DRAM I/F On-chip memory On-chip network Optimization Code generation Software Command scheduler Cache controller Network control Register Arithmetic units Hardware DRAM I/F On-chip memory On-chip network Details: https://projects.preferred.jp/en/mn-core/
18. 18 MN-Core™ Series Low-power AI processors (mainly for training) MN-Core Manufactured in TSMC 12nm process First operated in 2020 MN-Core 2 Manufactured in TSMC 7nm First operated in 2023 PFN and Kobe University are co-developing MN-Core series of highly energy-efficient AI processors to provide fast and vast computing power required for training foundation models and other AI models. 1st generation 2nd generation AI Chips Details: https://projects.preferred.jp/en/mn-core/
19. 19 MN-Core™ Series Low-power AI processor (mainly for generative AI inference) AI Chips Details: https://mn-core.com Currently under development, MN-Core L1000 features the highly energy-efficient MN-Core™ architecture and distributed 3D-stacked memory. PFN’s goal is to achieve a maximum of ten-fold increase in processing speed for generative AI inference compared with conventional processors. MN-Core L1000 Low-power processor for generative AI Launch planned in 2026
20. 20 In-House Supercomputers PFN combines advanced hardware and software technologies to build and operate its own computer clusters (supercomputers) for powerful and efficient computing required for AI development. Currently PFN provides the computational resources for partner companies as well as in-house projects. Computing Infrastructure GPU cluster MN-3 MN-Core™-powered MN-Server 2 MN-Core 2-powered Details: https://projects.preferred.jp/en/supercomputers/
21. 21 Energy-Efficient Computing Infrastructure PFN pursues highly energy-efficient computing infrastructure. Powered by PFN’s own AI chip MN-Core™ (first generation), MN-3 has topped the Green500 ranking of the world’s most energy-efficient supercomputers three times in June 2020, June 2021 and November 2021. Computing Infrastructure Jun. 2021 No. 1 Jun. 2020 No. 1 Nov. 2021 No. 1 Details: https://projects.preferred.jp/en/supercomputers/
22. 22 Cloud-based AI Computing Service Preferred Computing Platform™ (PFCP™) PFN launched PFCP in October 2024 to provide PFN’s AI-optimized computing infrastructure to AI developers MN-Server 2 specifications Accelerators: 8 MN-Core 2 chips (FP64 96TFlops, FP32 392TFlops, TF32 784TFlops, TF16 3.1PFlops) CPU: Intel® Xeon® Platinum 8480 + (2.0GHz) 2 processors, total of 112 cores Double-precision (FP64) performance: 8,960 GFlops Memory: 1,024GiB Storage: 960GB of System SSD and 15.3TB of operation SSD Inter-node network: 100Gbps Ethernet x4 Operating system and software Provides container image compatible with MN-Core 2 that runs on a Kubernetes pod Charges (excl. tax) Monthly charge: 10,000 yen Monthly exclusive use of MN-Server 2: 1.7 million yen/server Computing Infrastructure
23. 23 Generative AI foundation models PFN’s Multimodal Foundation Model: The Vision PFN plans to combine its multimodal foundation model* with foundation models for specialized data such as sensor values, molecular structures and genome. Under the plan, PFN aims to make the combined models to serve as a social and industrial infrastructure that solves complex problems that text-only large language models cannot. Multimodal Foundation Model Chat Text Photos + Special -ized model + Special -ized model + Special -ized model + Special -ized model Videos 3D Robot sensors Molecular structures Genome Plant sensors Training data Services Specialized applications *Multimodal foundation model: As opposed to large language models that are trained with text data only, a multimodal foundation model is trained with data in multiple modalities such as text, images and audio, which allows it to serve as a “common sense” for computers to understand the real world.
24. 24 PLaMo™ PFN Group’s multimodal foundation model PLaMo™ (“plah-mo”) is a generative AI foundation model developed fully from scratch by PFN’s group company Preferred Elements. Trained with high-quality data without using any existing large language models, PLaMo has achieved one of the highest scores on Japanese-language performance benchmarks. Generative AI foundation models • PLaMo Prime commercially launched in December 2024 • Small language model PLaMo Lite currently provided for edge use • Planning to launch task-specific models for finance, medicine, etc. World-class Japanese-language performance Japan-made, full-scratch model Easy to deploy through API • Scores higher than GPT-4 on major Japanese-language benchmarks* • Accuracy currently improving • Excels in Japanese-English translations *Common ”Jaster” benchmark for Japanese-language performance • Pre-trained and instruction- trained from scratch with unique architecture and training data • Not based on any external model: No restrictions due to external licenses or unclear sources • Provided through the cloud and API, easy to implement with simple re-coding • Also able to deploy in user’s closed, on-premise environment
25. 25 PLaMo™ Beta World’s top-level Japanese-language performance Details: https://www.pelements.jp/#product Generative AI foundation models
26. 26 PLaMo™ Lite Small language model for edge devices • 1 billion-parameter small language model that runs on edge devices • Incorporates technology derived from PLaMo-100B • High Japanese-language performance compared with similar-sized models Generative AI foundation models (Japanese-language questions) (Japanese-English translations)
27. 27 Solutions and Products Solutions and Products PFN combines its vertically-integrated AI expertise with specialized domain knowledge to develop solutions and products across multiple industries. Plant/Factory automation Entertainment Robots Retail/Logistics optimization Healthcare Drug discovery Materials discovery Education
28. Preferred Networks Solutions and Products (Selected)
29. 29 PreferredAI™ Suite of generative AI-powered products and services PFN provides products and services packaged for general business uses as PreferredAI™, as well as generative AI-powered solutions tailored for specific customer needs. Solutions and Products Support for aptitude assessment through AI avatar interactions, ensuring consistent talent acquisition. Smartly classify vast amounts of text from reviews and surveys to uncover valuable insights. PreferredAI Talent Scouter PreferredAI Work Suite PreferredAI Localization PreferredAI Insight Scan Eliminate the risk of oversights with automated slide reviews. PreferredAI Slide Review AI agent supports day-to-day operations using company-internal data. Generates consistent translations that takes account of glossary, context and culture. NEW NEW
30. 30 Plant Automation System Butadiene extraction unit Crude distillation unit Solutions and Products PFN and ENEOS co-developed AI system for automatic operation of a butadiene extraction unit and crude distillation unit in the ENEOS Kawasaki Refinery. Currently in continuous operation, the system achieved more economic, stable and efficient operation compared with manual operation. The two companies plan to deploy the system to other ENEOS refineries and provide it to external parties. The AI system successfully controls the stability even under external disturbances by maintaining the key operational value close to the target value. Successful AI-based automation achieved in December 12, continuous autonomous operation began in January 2023. • Input sensors: 363 • Controlled factors: 13 • Valves: 9 More details: https://www.preferred.jp/en/news/pr2 0230807/ Started world’s first AI-based autonomous operation in May 2024. Stabilizes fluctuations resulting from crude oil switching. • Input sensors: 930 • Controlled factors: 24 • Valves: 13 More details: https://www.preferred.jp/en/news/pr20 240524/ Problems to solve • Years of experience required for plant operation • Labor shortage for 24/7 operation Manufacturing
31. 31 Preferred Networks Visual Inspection Powered by PFN’s proprietary deep learning model, Preferred Networks Visual Inspection is a highly accurate, flexible and cost-effective software for manufacturers. Its system can be trained with only 100 images with “good” and “bad” labels without detailed annotations. Category Data Detected as non-defective Detected as defective Non- defective 850 850 0 Defective 128 0 128 10-minute DAGM2007 Class 1 test at 2:8 ratio for training:evaluation Zero overdetection, zero missed detection on benchmark dataset For minimal false negatives Human detection when in doubt Adjustable threshold 1. PFN’s proprietary deep learning model 2. Only 100 images required for training 3. Simple annotation 4. Visualizes defects 5. Intuitive GUI control 6. Flexible and quick system installation Problems to solve • Scarce training data for defects • Manual annotation required in conventional methods Solutions and Products Manufacturing
32. 32 Matlantis™ Universal atomistic simulator for materials discovery Co-developed by PFN and ENEOS, Matlantis™ performs simulation of innovative new materials for sustainable future including batteries, semiconductors, catalysts for e-fuel and lubricants over 10 thousand times faster than the conventional DFT method. Matlantis is currently provided by the joint venture Preferred Computational Chemistry (PFCC) as a cloud-based service to more than 100 companies and organizations. Problems to solve • Limited sustainability of conventional materials • Vast time and cost required for testing new materials Catalyst Battery Semiconductors Alloy Lubricant Ceramic Absorbent Separation membrane Supports combinations of 96 elements in various forms of materials including molecules and crystals Simulates new materials’ properties on web browser over 10,000 times faster than conventional method Supports discovery of a range of material categories that contribute to sustainability Details: https://matlantis.com/ 2,264 years would be required for 1 GPU to generate the training data with more than 59 million structures The neural network potential powering Matlantis has been developed using National Institute of Advanced Industrial Science and Technology’s AI Bridging Cloud Infrastructure (ABCI) in addition to PFN’s in-house supercomputers. Materials discovery Solutions and Products
33. 33 AI-Driven Drug Discovery PFN uses its AI technologies and computing infrastructure to accelerate discovery of pharmaceutical lead compounds, which would require a vast amount of time and cost using conventional technologies. Generated over 1 million compound structures on computer Optimized candidates with Optuna™, synthesized 13 compounds in lab Out of 13 shortlisted compounds, 7 showed inhibitory activity against viral replication Problems to solve • 10+ years and millions of dollars required for drug development • Less than 1/20,000 success rate for pharmaceutical compounds Joint Research with Kyoto Pharmaceutical University (KPU) Drug discovery Solutions and Products In a joint research project with KPU in 2021, PFN discovered multiple compounds that inhibit viral replication of SARS-CoV-2 using its AI technologies. (conceptual graphic) P-FEP: Calculation for drug discovery P-FEP is PFN’s service for pharmaceutical research, in which PFN uses its own supercomputer to perform relative binding free-energy perturbation to shortlist promising low-molecular compounds for new drugs. P-FEP’s prediction accuracy based on known FEP benchmarks J.Am. Chem. Soc. 2015, 137, 2695 J.Chem. Inf. Model. 2020, 60, 5457 Generative model Details: https://www.preferred.jp/en/news/pr20210906/
34. 34 Sebum RNA Analysis PFN combined its machine learning technology with Kao’s technology for monitoring RNAs in skin surface lipids (SSLs) to build an AI model that predicts skin conditions. The combined technologies are currently used for Kao’s individualized skin analysis service. PFN and Kao, in collaboration with Juntendo University, also conducted research that showed that the technology may help doctors diagnose Parkinson’s disease early. Juntendo-Kao-PFN joint research (September 2021) This graph shows the ROC curve of sebum RNA test results for Parkinson’s disease patients versus healthy controls. The closer the graph plot is to the left top point, the more accurate the test is. The area under the curve was 0.793 (1 for 100% accuracy) for prediction based on sebum RNA, age and sex. The score improved to 0.806 with severity added as a regressor. Details: https://www.preferred.jp/en/news/pr20210921/ Problems to solve • Expensive equipment required for testing skin conditions • Special invasive testing is required for diagnosing Parkinson’s disease Healthcare Solutions and Products Low accuracy ↘ ↖High accuracy
35. 35 PFN 3D Scan 3D scanning service Launched in June 2022, PFN 3D Scan allows users to create 3D models that are faithful in appearance to the actual objects. PFN 3D Scan takes advantage of a deep learning technique that reconstructs free-perspective 3D graphics from 2D photographs taken from multiple angles. The PFN 3D Scan system can reconstruct transparent, metallic or black objects which were previously hard to 3D-scan. PFN has scanned nearly 30,000 objects for use in virtual reality, e-commerce and more. Problems to solve • Time and effort for building 3D models • Difficulty to 3D-scan transparent, metallic and black objects https://youtu.be/RX4j6wxWev8 Details: https://pfn3d.com/ Objects scanned using PFN 3D Scan Solutions and Products Creative
36. 36 PFN has applied the technique used in PFN 3D Scan for volumetric scanning in which the system captures motions such as sports and performing arts in 3D space in any location without a dedicated studio and reconstructs the scenes as an animated 3D graphic that moves over time. PFN expects the technology can be applied to content production and more. Details: https://pfn3d.com/4d/ https://youtu.be/6lnbwFLIEjU https://youtu.be/xbngQWtmtQ8 Examples of animated graphics created with PFN 4D Scan The animation faithfully reconstructs reflections and movement of the water, also allowing viewers to pause and look at the objects from free perspectives. The background buildings are also reconstructed as 3D objects and the movements of the fabric are also reconstructed faithfully to the physical origin. Acquarium Dancing PFN 4D Scan Volumetric scanning system Solutions and Products Creative Problems to solve • Green screen required for volumetric scan • Tedious manual editing also required