Dmytro Mishkin
@ducha-aiki.bsky.social
Marrying classical CV and Deep Learning. I do things, which work, rather than being novel, but not working.
http://dmytro.ai
http://dmytro.ai
SAM 3D: 3Dfy Anything in Images
SAM 3D Team et al?
tl;dr: in title. 8-stage training, dataset, human labeling. Do not read tl;dr, read whole paper
arxiv.org/abs/2511.16624
SAM 3D Team et al?
tl;dr: in title. 8-stage training, dataset, human labeling. Do not read tl;dr, read whole paper
arxiv.org/abs/2511.16624
November 21, 2025 at 1:00 PM
SAM 3D: 3Dfy Anything in Images
SAM 3D Team et al?
tl;dr: in title. 8-stage training, dataset, human labeling. Do not read tl;dr, read whole paper
arxiv.org/abs/2511.16624
SAM 3D Team et al?
tl;dr: in title. 8-stage training, dataset, human labeling. Do not read tl;dr, read whole paper
arxiv.org/abs/2511.16624
POMA-3D: The Point Map Way to 3D Scene Understanding
Weixun Luo, Ranran Huang, Junpeng Jing, Krystian Mikolajczyk
tl;dr: in title + dataset.
arxiv.org/abs/2511.16567
Weixun Luo, Ranran Huang, Junpeng Jing, Krystian Mikolajczyk
tl;dr: in title + dataset.
arxiv.org/abs/2511.16567
November 21, 2025 at 12:53 PM
POMA-3D: The Point Map Way to 3D Scene Understanding
Weixun Luo, Ranran Huang, Junpeng Jing, Krystian Mikolajczyk
tl;dr: in title + dataset.
arxiv.org/abs/2511.16567
Weixun Luo, Ranran Huang, Junpeng Jing, Krystian Mikolajczyk
tl;dr: in title + dataset.
arxiv.org/abs/2511.16567
PartUV: Part-Based UV Unwrapping of 3D Meshes
Zhaoning Wang, Xinyue Wei, Ruoxi Shi, Xiaoshuai Zhang, Hao Su, Minghua Liu
arxiv.org/abs/2511.16659
tl;dr: in-title, start from learning-based split -> algorithmic finish
Zhaoning Wang, Xinyue Wei, Ruoxi Shi, Xiaoshuai Zhang, Hao Su, Minghua Liu
arxiv.org/abs/2511.16659
tl;dr: in-title, start from learning-based split -> algorithmic finish
November 21, 2025 at 12:49 PM
PartUV: Part-Based UV Unwrapping of 3D Meshes
Zhaoning Wang, Xinyue Wei, Ruoxi Shi, Xiaoshuai Zhang, Hao Su, Minghua Liu
arxiv.org/abs/2511.16659
tl;dr: in-title, start from learning-based split -> algorithmic finish
Zhaoning Wang, Xinyue Wei, Ruoxi Shi, Xiaoshuai Zhang, Hao Su, Minghua Liu
arxiv.org/abs/2511.16659
tl;dr: in-title, start from learning-based split -> algorithmic finish
Reposted by Dmytro Mishkin
@ducha-aiki.bsky.social you may be interested in this, Viktor implemented covariance weighted RANSAC, should hopefully be in poselib main pretty soon (already available here: github.com/PoseLib/Pose...)
PoseLib/test_cov.py at two_view_covariance · PoseLib/PoseLib
Minimal solvers for calibrated camera pose estimation - PoseLib/PoseLib
github.com
November 20, 2025 at 9:28 AM
@ducha-aiki.bsky.social you may be interested in this, Viktor implemented covariance weighted RANSAC, should hopefully be in poselib main pretty soon (already available here: github.com/PoseLib/Pose...)
Reposted by Dmytro Mishkin
RoMa v2 is now out! (github.com/Parskatt/rom..., arxiv.org/abs/2511.15706)
Here are the main improvements we made since RoMa:
Here are the main improvements we made since RoMa:
November 20, 2025 at 9:25 AM
RoMa v2 is now out! (github.com/Parskatt/rom..., arxiv.org/abs/2511.15706)
Here are the main improvements we made since RoMa:
Here are the main improvements we made since RoMa:
RoMa v2: Harder Better Faster Denser Feature Matching
@parskatt.bsky.social et 11 al.
tl;dr: in title.
Predict covariance per-pixel, more datasets, use DINOv3, adjust architecture.
arxiv.org/abs/2511.15706
@parskatt.bsky.social et 11 al.
tl;dr: in title.
Predict covariance per-pixel, more datasets, use DINOv3, adjust architecture.
arxiv.org/abs/2511.15706
November 20, 2025 at 9:08 AM
RoMa v2: Harder Better Faster Denser Feature Matching
@parskatt.bsky.social et 11 al.
tl;dr: in title.
Predict covariance per-pixel, more datasets, use DINOv3, adjust architecture.
arxiv.org/abs/2511.15706
@parskatt.bsky.social et 11 al.
tl;dr: in title.
Predict covariance per-pixel, more datasets, use DINOv3, adjust architecture.
arxiv.org/abs/2511.15706
Reposted by Dmytro Mishkin
Pro-tip: if you send me an email asking for an internship, a PhD or a postdoc position, don't copy/paste an over-hyped summary of one of my papers you just asked ChatGPT to spit out.
Pro-tip#2: don't do that with other professors as well. It's not just me.
Pro-tip#2: don't do that with other professors as well. It's not just me.
November 19, 2025 at 8:24 AM
Pro-tip: if you send me an email asking for an internship, a PhD or a postdoc position, don't copy/paste an over-hyped summary of one of my papers you just asked ChatGPT to spit out.
Pro-tip#2: don't do that with other professors as well. It's not just me.
Pro-tip#2: don't do that with other professors as well. It's not just me.
Reposted by Dmytro Mishkin
Thanks @ducha-aiki.bsky.social for sharing our work, we truly appreciate it! More related links are listed below, and we welcome the community's feedback.
Project Page: depth-anything-3.github.io
Code: github.com/ByteDance-Se...
Hugging Face Demo: huggingface.co/spaces/depth...
Project Page: depth-anything-3.github.io
Code: github.com/ByteDance-Se...
Hugging Face Demo: huggingface.co/spaces/depth...
November 18, 2025 at 4:06 AM
Thanks @ducha-aiki.bsky.social for sharing our work, we truly appreciate it! More related links are listed below, and we welcome the community's feedback.
Project Page: depth-anything-3.github.io
Code: github.com/ByteDance-Se...
Hugging Face Demo: huggingface.co/spaces/depth...
Project Page: depth-anything-3.github.io
Code: github.com/ByteDance-Se...
Hugging Face Demo: huggingface.co/spaces/depth...
Depth Anything 3: Recovering the Visual Space from Any Views
Haotong Lin, Sili Chen, Junhao Liew, Donny Y. Chen, Zhenyu Li, Guang Shi, Jiashi Feng, Bingyi Kang
tl;dr: DINOv2+reshape for multiview,
joint DPT, synth teacher. Depth-ray output.
Simpler VGGT.
arxiv.org/abs/2511.10647
Haotong Lin, Sili Chen, Junhao Liew, Donny Y. Chen, Zhenyu Li, Guang Shi, Jiashi Feng, Bingyi Kang
tl;dr: DINOv2+reshape for multiview,
joint DPT, synth teacher. Depth-ray output.
Simpler VGGT.
arxiv.org/abs/2511.10647
November 17, 2025 at 4:45 PM
Depth Anything 3: Recovering the Visual Space from Any Views
Haotong Lin, Sili Chen, Junhao Liew, Donny Y. Chen, Zhenyu Li, Guang Shi, Jiashi Feng, Bingyi Kang
tl;dr: DINOv2+reshape for multiview,
joint DPT, synth teacher. Depth-ray output.
Simpler VGGT.
arxiv.org/abs/2511.10647
Haotong Lin, Sili Chen, Junhao Liew, Donny Y. Chen, Zhenyu Li, Guang Shi, Jiashi Feng, Bingyi Kang
tl;dr: DINOv2+reshape for multiview,
joint DPT, synth teacher. Depth-ray output.
Simpler VGGT.
arxiv.org/abs/2511.10647
Reposted by Dmytro Mishkin
Reviewers will be released upon acceptance of the manuscript.
November 14, 2025 at 7:13 AM
Reviewers will be released upon acceptance of the manuscript.
Reposted by Dmytro Mishkin
Following up on Monday’s discussion, I articulate a few concrete positions on archives, surveys, and position papers.
The DOI Directorate
Articulating a few concrete positions on archives, surveys, and position papers
www.argmin.net
November 12, 2025 at 3:30 PM
Following up on Monday’s discussion, I articulate a few concrete positions on archives, surveys, and position papers.
There are many single great idea people. While they looks super smart (and the idea is great), but often they are not as smart as they look.
That's a 10-15 years observation.
That's a 10-15 years observation.
November 11, 2025 at 7:33 PM
There are many single great idea people. While they looks super smart (and the idea is great), but often they are not as smart as they look.
That's a 10-15 years observation.
That's a 10-15 years observation.
Great thread about great paper figures
So, I like a couple of things.
1. When architecture information is combined with a Figure showing what actually happens with the data:
1. When architecture information is combined with a Figure showing what actually happens with the data:
November 11, 2025 at 1:57 PM
Great thread about great paper figures
Last week we launched IMC2025-Ongoing on
@kaggle.com
The dataset is exactly as in IMC2025, but the competition is on-going for a year, making it better for academic leaderboard and persistency.
kaggle.com/competitions...
1/2
@kaggle.com
The dataset is exactly as in IMC2025, but the competition is on-going for a year, making it better for academic leaderboard and persistency.
kaggle.com/competitions...
1/2
Dmytro Mishkin 🇺🇦 on X: "Last week we launched IMC2025-Ongoing on @kaggle The dataset is exactly as in IMC2025, but the competition is on-going for a year, making it better for academic leaderboard and persistency. https://t.co/ejmeN3Gh5B 1/2" / X
Last week we launched IMC2025-Ongoing on @kaggle The dataset is exactly as in IMC2025, but the competition is on-going for a year, making it better for academic leaderboard and persistency. https://t.co/ejmeN3Gh5B 1/2
x.com
November 10, 2025 at 8:32 PM
Last week we launched IMC2025-Ongoing on
@kaggle.com
The dataset is exactly as in IMC2025, but the competition is on-going for a year, making it better for academic leaderboard and persistency.
kaggle.com/competitions...
1/2
@kaggle.com
The dataset is exactly as in IMC2025, but the competition is on-going for a year, making it better for academic leaderboard and persistency.
kaggle.com/competitions...
1/2
Reposted by Dmytro Mishkin
Why are you in 🦋?
(I'm) For the same reason I downloaded:
1. WhatsApp
2. Signal, used only when living in 🇳🇱.
3. Telegram
(in such order) to be in contact with someone/someppl
The 🐦 still offers better UX for me. Nevertheless,
Kudos to all folks engaged on making a clear blue sky 💪🏼
(I'm) For the same reason I downloaded:
1. WhatsApp
2. Signal, used only when living in 🇳🇱.
3. Telegram
(in such order) to be in contact with someone/someppl
The 🐦 still offers better UX for me. Nevertheless,
Kudos to all folks engaged on making a clear blue sky 💪🏼
Bsky is great to following people who are here, but bad at serendipity, which is the most important point for me.
Ofc, that’s just me
Ofc, that’s just me
November 9, 2025 at 5:52 AM
Why are you in 🦋?
(I'm) For the same reason I downloaded:
1. WhatsApp
2. Signal, used only when living in 🇳🇱.
3. Telegram
(in such order) to be in contact with someone/someppl
The 🐦 still offers better UX for me. Nevertheless,
Kudos to all folks engaged on making a clear blue sky 💪🏼
(I'm) For the same reason I downloaded:
1. WhatsApp
2. Signal, used only when living in 🇳🇱.
3. Telegram
(in such order) to be in contact with someone/someppl
The 🐦 still offers better UX for me. Nevertheless,
Kudos to all folks engaged on making a clear blue sky 💪🏼
Reposted by Dmytro Mishkin
Come work with me @slu-official.bsky.social! The Department of Computer Science is running an open rank search with an emphasis on AI and Cybersecurity (although we’ll consider other areas as well for extraordinary candidates). slu.wd5.myworkdayjobs.com/en-US/Career...
Open Rank Professor - Computer Science
Who is Saint Louis University? Founded in 1818, Saint Louis University is one of the nation’s oldest and most prestigious Catholic universities. SLU, which also has a campus in Madrid, Spain, is recog...
slu.wd5.myworkdayjobs.com
November 6, 2025 at 7:59 PM
Come work with me @slu-official.bsky.social! The Department of Computer Science is running an open rank search with an emphasis on AI and Cybersecurity (although we’ll consider other areas as well for extraordinary candidates). slu.wd5.myworkdayjobs.com/en-US/Career...
Reposted by Dmytro Mishkin
🧍♀️ Introducing Anny: an open, interpretable, and differentiable human body model for all ages.
Grounded in anthropometric data (MakeHuman) & WHO stats, Anny offers:
🧠 Interpretable shape control
👶👩🦳 Unified from infants to elders
🧩 Versatile for fitting, synthesis & HMR
🌍 Open under Apache 2.0
Grounded in anthropometric data (MakeHuman) & WHO stats, Anny offers:
🧠 Interpretable shape control
👶👩🦳 Unified from infants to elders
🧩 Versatile for fitting, synthesis & HMR
🌍 Open under Apache 2.0
November 6, 2025 at 4:37 PM
🧍♀️ Introducing Anny: an open, interpretable, and differentiable human body model for all ages.
Grounded in anthropometric data (MakeHuman) & WHO stats, Anny offers:
🧠 Interpretable shape control
👶👩🦳 Unified from infants to elders
🧩 Versatile for fitting, synthesis & HMR
🌍 Open under Apache 2.0
Grounded in anthropometric data (MakeHuman) & WHO stats, Anny offers:
🧠 Interpretable shape control
👶👩🦳 Unified from infants to elders
🧩 Versatile for fitting, synthesis & HMR
🌍 Open under Apache 2.0
Reposted by Dmytro Mishkin
Code is out. We provide the pre-trained ACE-G model, as well as code to replicate the paper results, and to evaluate ACE-G on new scenes. #ICCV2025
github.com/nianticspati...
github.com/nianticspati...
November 5, 2025 at 2:39 PM
Code is out. We provide the pre-trained ACE-G model, as well as code to replicate the paper results, and to evaluate ACE-G on new scenes. #ICCV2025
github.com/nianticspati...
github.com/nianticspati...
Between yesterday and today, there has been submitted at least 60 CVPR Workshop proposals.
November 3, 2025 at 11:37 PM
Between yesterday and today, there has been submitted at least 60 CVPR Workshop proposals.
Reposted by Dmytro Mishkin
Hello, a PSA. Arxiv is the server that many academics use to discuss their work before it is peer reviewed. It is widely viewed as a blessing.
November 3, 2025 at 9:41 PM
Hello, a PSA. Arxiv is the server that many academics use to discuss their work before it is peer reviewed. It is widely viewed as a blessing.
Great overview of the workshops and tutorials.
My favorites:
1) CAD representation
2) synthetic data to help city-scale reconstruction
3) trends in 3D vision
4) visual chain-of-thoughts?
My favorites:
1) CAD representation
2) synthetic data to help city-scale reconstruction
3) trends in 3D vision
4) visual chain-of-thoughts?
November 3, 2025 at 1:06 PM
Great overview of the workshops and tutorials.
My favorites:
1) CAD representation
2) synthetic data to help city-scale reconstruction
3) trends in 3D vision
4) visual chain-of-thoughts?
My favorites:
1) CAD representation
2) synthetic data to help city-scale reconstruction
3) trends in 3D vision
4) visual chain-of-thoughts?
I HATE long @bsky.app name-handles.
Common, exclude them from the char limit, or whatever.
On twitter, when authors are on platform, it means MORE space. Here is means LESS space
It is easy to fit the paper name + authors (handles) + tl;dr + arXiv link on twitter, but almost impossible here :(
Common, exclude them from the char limit, or whatever.
On twitter, when authors are on platform, it means MORE space. Here is means LESS space
It is easy to fit the paper name + authors (handles) + tl;dr + arXiv link on twitter, but almost impossible here :(
November 3, 2025 at 1:00 PM
I HATE long @bsky.app name-handles.
Common, exclude them from the char limit, or whatever.
On twitter, when authors are on platform, it means MORE space. Here is means LESS space
It is easy to fit the paper name + authors (handles) + tl;dr + arXiv link on twitter, but almost impossible here :(
Common, exclude them from the char limit, or whatever.
On twitter, when authors are on platform, it means MORE space. Here is means LESS space
It is easy to fit the paper name + authors (handles) + tl;dr + arXiv link on twitter, but almost impossible here :(
Scaling Image Geo-Localization to Continent Level
Philipp Lindenberger
@pesarlin.bsky.social @janhosang.bsky.social Matteo Balice @marcpollefeys.bsky.social Simon Lynen, Eduard Trulls
tl;dr: combine ground+aerial to get cell-prototype. Acc@ 42Gb = ground [email protected]
arxiv.org/abs/2510.26795
Philipp Lindenberger
@pesarlin.bsky.social @janhosang.bsky.social Matteo Balice @marcpollefeys.bsky.social Simon Lynen, Eduard Trulls
tl;dr: combine ground+aerial to get cell-prototype. Acc@ 42Gb = ground [email protected]
arxiv.org/abs/2510.26795
November 3, 2025 at 12:56 PM
Scaling Image Geo-Localization to Continent Level
Philipp Lindenberger
@pesarlin.bsky.social @janhosang.bsky.social Matteo Balice @marcpollefeys.bsky.social Simon Lynen, Eduard Trulls
tl;dr: combine ground+aerial to get cell-prototype. Acc@ 42Gb = ground [email protected]
arxiv.org/abs/2510.26795
Philipp Lindenberger
@pesarlin.bsky.social @janhosang.bsky.social Matteo Balice @marcpollefeys.bsky.social Simon Lynen, Eduard Trulls
tl;dr: combine ground+aerial to get cell-prototype. Acc@ 42Gb = ground [email protected]
arxiv.org/abs/2510.26795
Instance-Level Composed Image Retrieval
@billpsomas.bsky.social George Retsinas @nikos-efth.bsky.social Panagiotis Filntisis,Yannis Avrithis, Petros Maragos, Ondrej Chum, @gtolias.bsky.social
tl;dr: condition-based retrieval (+dataset) - old photo/sunset/night/aerial/model arxiv.org/abs/2510.25387
@billpsomas.bsky.social George Retsinas @nikos-efth.bsky.social Panagiotis Filntisis,Yannis Avrithis, Petros Maragos, Ondrej Chum, @gtolias.bsky.social
tl;dr: condition-based retrieval (+dataset) - old photo/sunset/night/aerial/model arxiv.org/abs/2510.25387
November 3, 2025 at 12:53 PM
Instance-Level Composed Image Retrieval
@billpsomas.bsky.social George Retsinas @nikos-efth.bsky.social Panagiotis Filntisis,Yannis Avrithis, Petros Maragos, Ondrej Chum, @gtolias.bsky.social
tl;dr: condition-based retrieval (+dataset) - old photo/sunset/night/aerial/model arxiv.org/abs/2510.25387
@billpsomas.bsky.social George Retsinas @nikos-efth.bsky.social Panagiotis Filntisis,Yannis Avrithis, Petros Maragos, Ondrej Chum, @gtolias.bsky.social
tl;dr: condition-based retrieval (+dataset) - old photo/sunset/night/aerial/model arxiv.org/abs/2510.25387