Jump to content

Codec Avatars

From VR & AR Wiki

Codec Avatars is a research program at Meta Platforms' Reality Labs Research for photorealistic, real-time avatars of real people, intended for telepresence in virtual reality and augmented reality. Meta describes the technology as "metric telepresence" that aims to make remote social presence "indistinguishable from reality."[1] The work is based at the Reality Labs lab in Pittsburgh, Pennsylvania, which Yaser Sheikh has led since joining Facebook in 2015.[2]

The name comes from the system's encoder-decoder design: cameras and microphones on a headset encode the wearer's face, body and voice into a compact numeric code, which is transmitted and then decoded on the receiving side into a rendered likeness.[2] The underlying models are neural networks trained on multi-view capture data, beginning with the "deep appearance model" that Stephen Lombardi, Jason Saragih, Tomas Simon and Sheikh published at SIGGRAPH 2018.[3] Michael Abrash's Oculus Connect 5 keynote in 2018 showed results from "an early system known as codec avatars," and Facebook Reality Labs described the project in detail in March 2019.[4][2] A widely reported demonstration was a September 2023 episode of the Lex Fridman Podcast recorded with Mark Zuckerberg in which both participants appeared as Codec Avatars while wearing Meta Quest Pro headsets.[5]

As of September 2026, Codec Avatars remain a research technology and have not shipped as a consumer feature. Recent versions use 3D Gaussian representations, can be built from a phone scan, and can run on a standalone Meta Quest 3.[6][7][8] In August 2026, UploadVR reported that Meta's Horizon OS contains "Hologram Calling" frameworks that reference Codec Avatars and that Meta appears to be calling the feature "Hologram."[9]

Reviewed 20 September 2026. Lead definition, program history and dates, capture-rig and headset figures, all 13 papers (authors, venues, DOIs, page and article numbers), Ava-256 dataset details, the 2022-2026 demonstrations and press reports, citation template compatibility, links and categories. About review dates.

How it works

A Codec Avatar has two halves. On the sending side, an encoder network takes images from cameras mounted on the head-mounted display (and, in later work, audio) and estimates a low-dimensional code describing the wearer's current expression, gaze and pose. On the receiving side, a decoder network turns that code into geometry and appearance that a rendering engine can draw from any viewpoint, so the remote person can be looked at as if they were in the room.[2] Meta's 2019 description put it as a "unique code, a numeric representation of the state of a person's body and environment" that is transmitted and then decoded to display the sender's likeness and expression.[2]

Building the decoder requires a large person-specific dataset. The 2018 deep appearance model learned a joint representation of facial geometry and view-specific texture from a multi-view camera rig using a variational autoencoder; conditioning the texture on the viewing direction let the model reproduce view-dependent effects such as specular highlights and compensate for imperfect geometry, and the output could be drawn by a conventional real-time renderer for VR.[3] Facebook Reality Labs built two capture studios in Pittsburgh, one for faces and one for bodies, with hundreds of high-resolution cameras each recording at about 1 GB per second and a combined capture rate of roughly 180 gigabytes per second; one studio also has 1,700 microphones for reconstructing sound fields. A typical capture lasted around 15 minutes, and producing a finished avatar required several hours of capture followed by weeks of processing.[2] In a 2019 visit, CBS News reported a face rig of 180 cameras and 500 lights.[10] By 2022 the face rig in use had 171 cameras.[11]

Driving the avatar from a headset is a separate problem, because headset cameras see the face from oblique, partial angles that look nothing like the studio views the decoder was trained on. Wei et al. (2019) addressed this with a bidirectional system that used a "training headset" carrying extra cameras in favorable positions to establish correspondence between headset imagery and the avatar, then trained a real-time network for a "tracking headset" with a minimal camera set; self-supervised multi-view image translation found the correspondence between headset camera images and the avatar without manual annotation.[12][13] The prototype headset described in 2022 carried five cameras: two inside the headset viewing each eye and three outside viewing the lower face.[11] Bai et al. (2024) later trained a single "universal" facial encoder that generalizes to people it has never seen after a lightweight calibration, using a self-supervised cross-view reconstruction objective to cope with oblique views, headset placement and lighting variation.[14]

History

Origins (2014-2019)

Sheikh, a computer vision researcher at Carnegie Mellon University's Robotics Institute,[15] met Oculus Research chief scientist Abrash in the fall of 2014, joined Facebook in 2015 and led the Pittsburgh team from then on.[2] Meta dates the lab's work on lifelike avatars to 2015.[1] The first published Codec Avatars paper, "Deep appearance models for face rendering," appeared at SIGGRAPH 2018.[3] At Oculus Connect 5 that year, Abrash's keynote showed the results as "an early system known as codec avatars."[4]

On 12 March 2019 Facebook Reality Labs published a detailed account of the project, including the encoder-decoder concept, the Pittsburgh capture studios, headset-mounted camera prototypes, and a warning that the technology was "still years away" from consumer headsets. The post also discussed deepfake risk, which the team called an "existential threat" to the telepresence project, and the authentication and encryption measures under study.[2] CBS News visited the lab the same week; Sheikh told CBS that the eventual goal was to build an avatar from a few photos or a short video supplied by the user.[10] The headset-driving system was presented at SIGGRAPH 2019 in Los Angeles (28 July to 1 August 2019).[13]

Efficiency, bodies and phone scans (2020-2022)

Subsequent papers made the avatars cheaper to render and easier to create. Mixture of Volumetric Primitives (MVP, SIGGRAPH 2021) replaced the mesh-plus-texture decoder with a set of small volumetric primitives attached to a coarse mesh, which handles hair and other thin structures that triangle meshes model poorly while remaining fast enough for real time, unlike Neural Radiance Fields of the period.[16] Pixel Codec Avatars (CVPR 2021) moved most of the decoding to a per-pixel network that only computes visible pixels; the authors rendered five avatars at once in real time on a single Oculus Quest 2.[17] "Authentic volumetric avatars from a phone scan" (SIGGRAPH 2022) trained a universal prior on hundreds of multi-view captures and then specialized it to a new person from a neutral-expression phone scan using inverse rendering, removing the need for a studio session.[18]

Speaking at MIT in April 2022, Sheikh said the project had started "ten miracles away" from shipping and was then "five miracles away"; the same talk described a Quest 2 rendering five avatars in an empty scene at 50 frames per second.[11] In October 2022 he told UploadVR that the first real-time cross-continental Codec Avatar call had taken place in 2021, and that "three to four research miracles" stood between the lab and running the avatars on the on-board compute of the Quest Pro; creating a face-and-body avatar still required several hours in two scanning systems and at least four weeks of processing.[19]

Public demonstrations (2022-2023)

At Meta Connect 2022 on 11 October 2022, Reality Labs Research showed "Codec Avatars 2.0," which added non-verbal cues beyond basic gaze direction (raised eyebrows, squints, widened eyes, a scrunched nose) and let the user change the lighting on the avatar, plus "Instant Codec Avatars," a lower-fidelity version built from a smartphone scan of about 30 seconds of neutral expression and 90 seconds of varied expressions, followed by a few hours of processing. The lab also showed changeable virtual outfits.[20]

On 28 September 2023 Lex Fridman released episode 398 of his podcast, "Mark Zuckerberg: First Interview in the Metaverse," recorded with both participants appearing as Codec Avatars.[21] Both wore Quest Pro headsets, whose built-in eye and face tracking drove the avatars; the avatars themselves had been generated in the capture rig of more than 100 cameras. UploadVR reported that Fridman said he genuinely forgot the avatar he was looking at was not a real person.[5] Speaking at Nvidia's GTC conference in March 2024, Sheikh showed the actual recording setup: each side used a wired Quest Pro with five face and eye tracking sensors, and a workstation with four GeForce RTX 4090 graphics cards rendered the avatars.[22]

Meta chief technology officer Andrew Bosworth said in March 2023 that the company was "making progress" toward shipping the avatars but was "still years away." In January 2024 references to Codec Avatars were found in Quest firmware version 62.[23]

Gaussian representations and open datasets (2023-2025)

In December 2023 the lab introduced Relightable Gaussian Codec Avatars, which replaced volumetric primitives with 3D Gaussians and added a learnable radiance-transfer appearance model. The Gaussian geometry captured sub-millimeter detail such as hair strands and pores, the appearance model supported real-time relighting under point lights and environment maps, and explicit eye models gave gaze control; the paper, an oral at CVPR 2024, demonstrated real-time relighting on a tethered consumer VR headset.[6] URAvatar (SIGGRAPH Asia 2024) combined this with the phone-scan approach: a universal relightable Gaussian model trained on hundreds of multi-view scans under controlled point lights is fine-tuned to a phone scan taken under unknown lighting, yielding an avatar that can be relit in real time.[7]

On 12 June 2024 Meta released Ava-256, a dataset pairing high-resolution dome captures with headset captures of the same people, which Meta says enables "for the first time" research into large-scale end-to-end photorealistic telepresence: 256 subjects recorded in a 172-camera dome and again wearing a Quest Pro fitted with five additional infrared cameras, plus code and pretrained universal encoder and decoder models, under a CC BY-NC 4.0 license.[24][25] The lab hosted a full-day Codec Avatars workshop at CVPR 2024 in Seattle on 18 June 2024,[26] and the dataset paper, "Codec Avatar Studio," which also introduced the eight-modality Goliath-4 dataset, was published in the NeurIPS 2024 Datasets and Benchmarks track.[27]

Full-body work followed. Relightable Full-Body Gaussian Codec Avatars (SIGGRAPH 2025) modeled relighting under body articulation with zonal harmonics, a shadow network and deferred specular shading.[28] SqueezeMe (SIGGRAPH 2025) distilled full-body Gaussian avatars so that three could be animated and rendered simultaneously at 72 frames per second on a Meta Quest 3, running the decoder on the NPU of its Snapdragon XR2 Gen 2 chip and splatting on the GPU through Vulkan;[29] UploadVR noted that these avatars still had flat lighting and no dynamic relighting, and were captured with the large camera array rather than a phone.[8] Vid2Avatar-Pro (CVPR 2025) built animatable full-body avatars from ordinary monocular video using a universal prior learned from multi-view clothed-human captures.[30]

Toward a product (2025-2026)

According to his published profile, Sheikh was Director of Facebook Reality Labs in Pittsburgh from April 2018 and a Meta vice president from August 2023 until September 2025; in January 2026 he became chief executive of Sooth Labs, an AI forecasting startup he co-founded.[15][31][32] The lab's research continued: Large-scale Codec Avatars (LCA), accepted to CVPR 2026, is a full-body model trained at large scale that generalizes to new people in a feed-forward pass.[33]

UploadVR's August 2026 overview of Meta's next headset (code-named Phoenix) reported several signs of productization: December 2025 firmware strings describing how to set up and calibrate face tracking for an avatar, in-development Codec Avatar profiles of Meta employees visible through the Meta Horizon website in December 2025, "Hologram Calling" frameworks in Horizon OS with extensive references to Codec Avatars, and apparently AI-rendered "Hologram" product tour videos found in August 2026 builds of the Meta AI mobile app showing the avatars on Ray-Ban Display-like smart glasses. The same article noted that Meta's cancelled Quest Pro successor (code-named La Jolla) had reportedly been set to feature Codec Avatars.[9] On 16 September 2026, Road to VR relayed a report by The Information that Meta planned to show a "hologram-like" video call built on an advanced version of Codec Avatars during the Connect 2026 keynote (the event runs 23-24 September 2026), while noting it was uncertain whether this would be a commercial launch or another look at in-progress research.[34] UploadVR added on 18 September that the plan was for a full-body photorealistic hologram call during a live Phoenix demo, with Bosworth appearing as the hologram.[35] As of 20 September 2026 that event has not yet taken place.

Research

The Codec Avatars program has produced a series of peer-reviewed papers, mostly at SIGGRAPH, SIGGRAPH Asia and CVPR. The table lists the principal ones.

Year Paper Venue Contribution
2018 Deep appearance models for face rendering (Lombardi et al.)[3] SIGGRAPH 2018 (ACM TOG 37:4) Variational autoencoder of geometry and view-dependent texture from multi-view capture; the original codec avatar decoder
2019 VR facial animation via multiview image translation (Wei et al.)[12] SIGGRAPH 2019 (ACM TOG 38:4) Driving the avatar in real time from headset-mounted cameras via training and tracking headsets
2021 Mixture of volumetric primitives for efficient neural rendering (Lombardi et al.)[16] SIGGRAPH 2021 (ACM TOG 40:4) Volumetric primitives on a coarse mesh for hair and thin structures at real-time rates
2021 Pixel Codec Avatars (Ma et al.)[17] CVPR 2021 Per-pixel decoding; five avatars in real time on an Oculus Quest 2
2022 Authentic volumetric avatars from a phone scan (Cao et al.)[18] SIGGRAPH 2022 (ACM TOG 41:4) Universal prior specialized to a new person from a phone scan
2024 Relightable Gaussian Codec Avatars (Saito et al.)[6] CVPR 2024 (oral) 3D Gaussian geometry with learnable radiance transfer; real-time relighting on a tethered headset
2024 Universal Facial Encoding of Codec Avatars from VR Headsets (Bai et al.)[14] SIGGRAPH 2024 (ACM TOG 43:4) Encoder that generalizes to unseen users after light calibration
2024 Codec Avatar Studio (Martinez et al.)[27] NeurIPS 2024 Datasets and Benchmarks Ava-256 and Goliath-4 paired-capture datasets with code and models
2024 URAvatar (Li et al.)[7] SIGGRAPH Asia 2024 Relightable Gaussian avatar from a phone scan under unknown lighting
2025 Relightable Full-Body Gaussian Codec Avatars (Wang et al.)[28] SIGGRAPH 2025 Relightable full-body avatars including face and hands
2025 SqueezeMe (Iandola et al.)[29] SIGGRAPH 2025 Three full-body Gaussian avatars at 72 fps on a Meta Quest 3
2025 Vid2Avatar-Pro (Guo et al.)[30] CVPR 2025 Full-body avatar from monocular in-the-wild video via a universal prior
2026 Large-scale Codec Avatars (Li et al.)[33] CVPR 2026 Feed-forward full-body avatar model trained at large scale

Beyond the papers, Meta's Codec Avatars page lists open datasets and reference implementations released for telepresence research, including Multiface (facial recordings of 13 people), CT2Hair, InterHand2.6M, Re:InterHand, Eyeful, Sounding Bodies and PatternedClothing, and names generalization of universal priors to new identities, online encoder adaptation and the quality of clothing and hair as open problems.[1]

Applications in VR and AR

The stated purpose of Codec Avatars is remote communication that preserves eye contact, small changes of expression, posture and gesture, so that two people in different places can interact as if sharing a room.[1] Sheikh has framed this as a successor to the telephone and video conferencing,[10] and as a "grand challenge of the next decade" to make remote interaction indistinguishable from being in person.[11] The efficiency work is directed at multi-person use on standalone headsets: five faces on a Quest 2 in 2021 and three full bodies on a Quest 3 in 2025.[17][29] In an earlier interview with Lex Fridman, relayed by UploadVR in May 2022, Zuckerberg described a future in which a user might pick an "expressionist" avatar for casual games and a "realistic" one for work meetings, alongside the stylized avatars already used in social VR.[11]

Because the headset encoder depends on inward-facing cameras, the technology is tied to headsets with eye and face tracking such as the Quest Pro; a Quest 3 lacks those sensors.[5][23] The 2026 reporting described above suggests that Meta's current direction for the technology is video calling and telepresence across its apps and devices under the "Hologram" name.[9]

See also

References

  1. ↑ 1.0 1.1 1.2 1.3 "Immersive Telepresence with Lifelike Avatars". Meta. Meta Platforms. https://www.meta.com/emerging-tech/codec-avatars/. Retrieved 2026-09-20.
  2. ↑ 2.0 2.1 2.2 2.3 2.4 2.5 2.6 2.7 "Facebook is building the future of connection with lifelike avatars". Tech at Meta. Meta Platforms. 2019-03-12. https://tech.facebook.com/ar-vr/2019/03/codec-avatars-facebook-reality-labs/. Retrieved 2026-09-20.
  3. ↑ 3.0 3.1 3.2 3.3 Lombardi, Stephen; Saragih, Jason; Simon, Tomas; Sheikh, Yaser (2018). "Deep appearance models for face rendering". ACM Transactions on Graphics. Vol. 37, no. 4, Article 68. https://doi.org/10.1145/3197517.3201401. Retrieved 2026-09-20.
  4. ↑ 4.0 4.1 Jamie Feltham (2018-10-02). "Mixing Realities, True Haptics And Photorealistic Humans: 5 Big Takeaways From Michael Abrash's OC5 Keynote". UploadVR. https://www.uploadvr.com/abrash-2018-predictions-oc5/. Retrieved 2026-09-20.
  5. ↑ 5.0 5.1 5.2 David Heaney (2023-09-28). "Watch Zuckerberg Interviewed In VR With Photoreal Avatars". UploadVR. https://www.uploadvr.com/mark-zuckerberg-lex-fridman-interview-photorealistic-codec-avatars/. Retrieved 2026-09-20.
  6. ↑ 6.0 6.1 6.2 Saito, Shunsuke; Schwartz, Gabriel; Simon, Tomas; Li, Junxuan; Nam, Giljoo (2024). "Relightable Gaussian Codec Avatars". 2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), doi:10.1109/CVPR52733.2024.00021. pp. 130-141. https://openaccess.thecvf.com/content/CVPR2024/html/Saito_Relightable_Gaussian_Codec_Avatars_CVPR_2024_paper.html.
  7. ↑ 7.0 7.1 7.2 Li, Junxuan; Cao, Chen; Schwartz, Gabriel; Khirodkar, Rawal; Richardt, Christian; Simon, Tomas; Sheikh, Yaser; Saito, Shunsuke (2024). "URAvatar: Universal Relightable Gaussian Codec Avatars". SIGGRAPH Asia 2024 Conference Papers, doi:10.1145/3680528.3687653. https://arxiv.org/abs/2410.24223.
  8. ↑ 8.0 8.1 David Heaney (2025-08-04). "Meta Got 3 Full-Body Codec Avatars Running On Quest 3". UploadVR. https://www.uploadvr.com/meta-squeezeme-mobile-ready-distillation-of-gaussian-full-body-avatars/. Retrieved 2026-09-20.
  9. ↑ 9.0 9.1 9.2 Luna (2026-08-24). "Everything We Know (So Far) About Meta's Next Headset". UploadVR. https://www.uploadvr.com/phoenix-everything-we-know-so-far-meta-next-headset/. Retrieved 2026-09-20.
  10. ↑ 10.0 10.1 10.2 DeMarco Morgan (2019-03-13). "Facebook VR video chat: Inside the Facebook virtual reality lab building lifelike avatars". CBS News. https://www.cbsnews.com/news/facebook-vr-video-chat-social-media-platform-is-creating-lifelike-replicas-of-real-people/. Retrieved 2026-09-20.
  11. ↑ 11.0 11.1 11.2 11.3 11.4 David Heaney (2022-05-04). "Meta Research: Codec Avatars 2.0 Approach Complete Realism". UploadVR. https://www.uploadvr.com/codec-avatars-2-0-photorealism/. Retrieved 2026-09-20.
  12. ↑ 12.0 12.1 Wei, Shih-En; Saragih, Jason; Simon, Tomas; Harley, Adam W.; Lombardi, Stephen; Perdoch, Michal; Hypes, Alexander; Wang, Dawei; Badino, Hernan; Sheikh, Yaser (2019). "VR facial animation via multiview image translation". ACM Transactions on Graphics. Vol. 38, no. 4. https://doi.org/10.1145/3306346.3323030. Retrieved 2026-09-20.
  13. ↑ 13.0 13.1 "Virtual reality faces: animating precise, lifelike avatars for VR in real-time". Phys.org. Association for Computing Machinery. 2019-06-25. https://phys.org/news/2019-06-virtual-reality-facesanimating-precise-lifelike.html. Retrieved 2026-09-20.
  14. ↑ 14.0 14.1 Bai, Shaojie; Wang, Te-Li; Li, Chenghui; Venkatesh, Akshay; Simon, Tomas; Cao, Chen; Schwartz, Gabriel; Wrench, Ryan; Saragih, Jason; Sheikh, Yaser; Wei, Shih-En (2024). "Universal Facial Encoding of Codec Avatars from VR Headsets". ACM Transactions on Graphics. Vol. 43, no. 4, Article 93. https://doi.org/10.1145/3658234. Retrieved 2026-09-20.
  15. ↑ 15.0 15.1 "Yaser Sheikh". alphaXiv. https://www.alphaxiv.org/@yaser-sheikh. Retrieved 2026-09-20.
  16. ↑ 16.0 16.1 Lombardi, Stephen; Simon, Tomas; Schwartz, Gabriel; Zollhoefer, Michael; Sheikh, Yaser; Saragih, Jason (2021). "Mixture of volumetric primitives for efficient neural rendering". ACM Transactions on Graphics. Vol. 40, no. 4, Article 59. https://doi.org/10.1145/3450626.3459863. Retrieved 2026-09-20.
  17. ↑ 17.0 17.1 17.2 Ma, Shugao; Simon, Tomas; Saragih, Jason; Wang, Dawei; Li, Yuecheng; De la Torre, Fernando; Sheikh, Yaser (2021). "Pixel Codec Avatars". 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), doi:10.1109/CVPR46437.2021.00013. pp. 64-73. https://openaccess.thecvf.com/content/CVPR2021/html/Ma_Pixel_Codec_Avatars_CVPR_2021_paper.html.
  18. ↑ 18.0 18.1 Cao, Chen; Simon, Tomas; Kim, Jin Kyu; Schwartz, Gabriel; Zollhoefer, Michael; Saito, Shunsuke; Lombardi, Stephen; Wei, Shih-En; Belko, Danielle; Yu, Shoou-I; Sheikh, Yaser; Saragih, Jason (2022). "Authentic volumetric avatars from a phone scan". ACM Transactions on Graphics. Vol. 41, no. 4, Article 163. https://doi.org/10.1145/3528223.3530143. Retrieved 2026-09-20.
  19. ↑ Ian Hamilton (2022-10-11). "VR's Killer App May Be Full-Body Codec Avatars". UploadVR. https://www.uploadvr.com/vr-killer-app-avatar-telepresence/. Retrieved 2026-09-20.
  20. ↑ "Reality Labs Research at Meta Connect 2022". Tech at Meta. Meta Platforms. 2022-10-11. https://tech.facebook.com/ar-vr/2022/10/meta-research-reality-labs-connect-2022/. Retrieved 2026-09-20.
  21. ↑ Lex Fridman (2023-09-28). "#398 - Mark Zuckerberg: First Interview in the Metaverse". Lex Fridman Podcast. https://lexfridman.com/mark-zuckerberg-3/. Retrieved 2026-09-20.
  22. ↑ Tomislav Bezmalinovic (2024-03-26). "Mark Zuckerberg's impressive Codec Avatar was rendered on a powerful workstation". MIXED. https://mixed-news.com/en/mark-zuckerberg-lex-fridman-codec-avatars-workstation/. Retrieved 2026-09-20.
  23. ↑ 23.0 23.1 David Heaney and Luna (2024-01-24). "Is Meta Bringing Photoreal 'Codec Avatars' To Quest Soon?". UploadVR. https://www.uploadvr.com/meta-codec-avatars-might-be-coming-to-quest/. Retrieved 2026-09-20.
  24. ↑ "Ava-256 Dataset Datasheet". GitHub. Meta Reality Labs Research. https://github.com/facebookresearch/ava-256/blob/main/DATASHEET.md. Retrieved 2026-09-20.
  25. ↑ "Codec Avatars: Ava-256 Dataset". Meta. Meta Platforms. https://www.meta.com/emerging-tech/codec-avatars/ava256/. Retrieved 2026-09-20.
  26. ↑ "Codec Avatars Workshop @ CVPR'24". Codec Avatars Workshop. Meta. https://codec-avatars.github.io/cvpr24/. Retrieved 2026-09-20.
  27. ↑ 27.0 27.1 Martinez, Julieta; Kim, Emily; Romero, Javier; Bagautdinov, Timur; Saito, Shunsuke (2024). "Codec Avatar Studio: Paired Human Captures for Complete, Driveable, and Generalizable Avatars". Advances in Neural Information Processing Systems 37 (NeurIPS 2024), Datasets and Benchmarks Track. https://proceedings.neurips.cc/paper_files/paper/2024/hash/9712b78386cebdc3db7f1a48c2d20edb-Abstract-Datasets_and_Benchmarks_Track.html.
  28. ↑ 28.0 28.1 Wang, Shaofei; Simon, Tomas; Santesteban, Igor; Bagautdinov, Timur; Li, Junxuan (2025). "Relightable Full-Body Gaussian Codec Avatars". SIGGRAPH 2025 Conference Papers, doi:10.1145/3721238.3730739. https://arxiv.org/abs/2501.14726.
  29. ↑ 29.0 29.1 29.2 Forrest Iandola, Stanislav Pidhorskyi, Igor Santesteban, Divam Gupta, Anuj Pahuja, Nemanja Bartolovic, Frank Yu, Emanuel Garbin, Tomas Simon, Shunsuke Saito (2025-06-27). "SqueezeMe: Mobile-Ready Distillation of Gaussian Full-Body Avatars". arXiv. https://arxiv.org/abs/2412.15171. Retrieved 2026-09-20.
  30. ↑ 30.0 30.1 Guo, Chen; Li, Junxuan; Kant, Yash; Sheikh, Yaser; Saito, Shunsuke; Cao, Chen (2025). "Vid2Avatar-Pro: Authentic Avatar from Videos in the Wild via Universal Prior". 2025 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). pp. 5559-5570. https://openaccess.thecvf.com/content/CVPR2025/html/Guo_Vid2Avatar-Pro_Authentic_Avatar_from_Videos_in_the_Wild_via_Universal_CVPR_2025_paper.html.
  31. ↑ "Sooth Labs - World Model for Global Foresight". Sooth Labs. https://soothlabs.com/. Retrieved 2026-09-20.
  32. ↑ "Why We Invested: Sooth Labs". Ardent Venture Partners. 2026-06-26. https://ardent.vc/blog-posts/why-we-invested-sooth-labs. Retrieved 2026-09-20.
  33. ↑ 33.0 33.1 Junxuan Li, Rawal Khirodkar, Chengan He, Zhongshi Jiang, Giljoo Nam, Lingchen Yang, Jihyun Lee, Egor Zakharov, Zhaoen Su, Rinat Abdrashitov, Yuan Dong, Julieta Martinez, Kai Li, Qingyang Tan, Takaaki Shiratori, Matthew Hu, Peihong Guo, Xuhua Huang, Ariyan Zarei, Marco Pesavento, Yichen Xu, He Wen, Teng Deng, Wyatt Borsos, Anjali Thakrar, Jean-Charles Bazin, Carsten Stoll, Ginés Hidalgo, James Booth, Lucy Wang, Xiaowen Ma, Yu Rong, Sairanjith Thalanki, Chen Cao, Christian Häne, Abhishek Kar, Sofien Bouaziz, Jason Saragih, Yaser Sheikh, Shunsuke Saito (2026-04-02). "Large-scale Codec Avatars: The Unreasonable Effectiveness of Large-scale Avatar Pretraining". arXiv. https://arxiv.org/abs/2604.02320. Retrieved 2026-09-20.
  34. ↑ Scott Hayden (2026-09-16). "Meta Reportedly Unveiling Next-gen MR Headset at Connect, Hologram-like Call Experience". Road to VR. https://roadtovr.com/meta-project-phoenix-connect-unveiling-codec-avatar/. Retrieved 2026-09-20.
  35. ↑ Luna (2026-09-18). "Meta Could Demo Next-Gen Headset At Connect With Full-Body Codec Avatars". UploadVR. https://www.uploadvr.com/meta-could-demo-phoenix-headset-at-connect-with-full-body-codec-avatars/. Retrieved 2026-09-20.
Retrieved from "https://vrarwiki.com/index.php?title=Codec_Avatars&oldid=42244"