Kirill Mazur
@makezur

PhD student at Dyson Robotics Lab @ Imperial College London

 London, UK
Joined May 2020
15 Photos and videos
AgentSTAR is an agentic method for both reconstruction and tracking of articulated objects from monocular videos. We can now track through rapid motion, severe occlusion and thin objects! 

There’s no spoon but we can still track it 🥄
10
45
430
78,844
Wrote done my thoughts on what's fundamentally new for vision when using agents --- obviously, apart from some hard problems that have now been nearly solved :)
Article

Agentic Optimisation

It’s really easy to get fooled when trying out these agents. A lot of the demos we see now, framed as “Astra did this” or “Astra did that,” are actually examples of agents autonomously using tools we

1
9
54
8,721
Christopher Fraser@RealCFraser
Sep 15
I wonder if something like this could be used for Star Wars fan films to track 3d lightsabers instead of using roto🤔
1
18
1,222
AgentSTAR is an agentic method for both reconstruction and tracking of articulated objects from monocular videos. We can now track through rapid motion, severe occlusion and thin objects! 

There’s no spoon but we can still track it 🥄
10
45
430
78,844
I particularly like this example, where we show our reconstruction alongside the ground-truth hand poses -- it gives a much better sense of the 3D tracking accuracy
1
2
502
For more visuals and details check out our project page, paper and code release! Real2Sim for dynamic scenes is definitely within reach now joint w/ @n_karaev Matthew Chang @JitendraMalikCV @notmahi Project page: agenticstar.github.io Code: github.com/makezur/agenticST…

AgentSTAR: Agentic Shape Tracking and Reconstruction from Monocular Videos

AgentSTAR: Agentic Shape Tracking and Reconstruction from Monocular Videos. Project page.

agenticstar.github.io
1
2
24
2,366
In my mind, this line of thinking was popularised by Real2Code and SceneScript, but can now finally be taken to the extreme, where code is not only the representation but also what gets the tracking done
15
1,825
Kirill Mazur retweeted
Nicholas Fry@deepfry_n
Sep 7
I'll be presenting our work at #ECCV2026 in Malmö! 🇸🇪 Find me at the poster and demo sessions: Poster 📅 Thu 10 Sep, 10:30 📍 ExHall #272 Demo 📅 Sat 12 Sep, 8:00 📍 ExHall Also at NeuSLAM Workshop 🤖: Tue 8 Sep, 14:00, Malmö Arena Hotel
Nicholas Fry@deepfry_n
Jun 1
(1/3) I am happy to share our work on Triangle Splatting SLAM! We show the first RGB-D SLAM to use differentiable triangles as a 3D map representation. Our method enables online mesh-based deformations and collision checking via on-the-fly Delaunay triangulation.
4
20
1,659
Kirill Mazur retweeted
Andrew Davison
@AjdDavison
Jun 22
Just announced: #CVPR2026 Best Demo Award for KV-Tracker. Congrats to Marwan and co-authors. And thanks to @deepfry_n, @shinjeong99 for massive help running the demo. That's 4 CVPR Best Demo Awards/Honourable Mentions in 3 years for our lab at Imperial! Proud of our demo culture.
Marwan Taher@marwan_ptr
19 Dec 2025
How can we run reconstruction models like π³ and Depth Anything 3 in real-time? We present KV-Tracker, a training-free approach, for real-time tracking of scenes and objects. Achieving up to 30 FPS! With @alzugarayign, @makezur, @XinKong_IC and @AjdDavison
2
20
217
23,756
Kirill Mazur retweeted
Marwan Taher@marwan_ptr
Jun 6
Wanna see a real-time demo of KV-Tracker? Stop by in ExHall F today 11:45 - 13:45! #CVPR2026
2
12
54
4,961
How to build complete 4D reconstructions from videos? Swing by 4DPM oral presentation today: Mile High Ballroom 3A - 4A, 13:00 #CVPR2026
4
8
107
11,022
code is released too: github.com/makezur/4D_PM project page: makezur.github.io/4DPM/

GitHub - makezur/4D_PM: [CVPR 2026 Oral] 4D Primitive-Mâché: Glueing Primitives for Persistent 4D...

[CVPR 2026 Oral] 4D Primitive-Mâché: Glueing Primitives for Persistent 4D Scene Reconstruction - makezur/4D_PM

github.com
3
681
4DPM got selected for an oral presentation at #CVPR2026 🤠
Kirill Mazur
@makezur
18 Dec 2025
Introducing 4D Primitive-Mâché (4DPM), a new method for replayable 4D reconstruction from monocular videos. We split dynamic scenes into 3D primitives and recover their motion. 4DPM can infer object positions even after they leave view. Joint work with @marwan_ptr @AjdDavison
3
5
65
7,204
accepted to #CVPR2026 😎
Kirill Mazur
@makezur
18 Dec 2025
Introducing 4D Primitive-Mâché (4DPM), a new method for replayable 4D reconstruction from monocular videos. We split dynamic scenes into 3D primitives and recover their motion. 4DPM can infer object positions even after they leave view. Joint work with @marwan_ptr @AjdDavison
1
6
87
7,168
project page: makezur.github.io/4DPM/

4DPM — 4D Primitive-Mâché

Glueing Primitives for Persistent 4D Scene Reconstruction — paper, video and results.

makezur.github.io
215
Kirill Mazur retweeted
Zhenjun Zhao@zhenjun_zhao
20 Dec 2025
4D Primitive-Mâché: Glueing Primitives for Persistent 4D Scene Reconstruction @makezur, @marwan_ptr, @AjdDavison tl;dr: improve SuperPrimitive; scenes->3D primitives->sparse per-primitive SE(3) poses (+2D correspondences)->per-pixel motion fields arxiv.org/abs/2512.16564
7
33
1,935
Kirill Mazur retweeted
Marwan Taher@marwan_ptr
19 Dec 2025
How can we run reconstruction models like π³ and Depth Anything 3 in real-time? We present KV-Tracker, a training-free approach, for real-time tracking of scenes and objects. Achieving up to 30 FPS! With @alzugarayign, @makezur, @XinKong_IC and @AjdDavison
11
99
741
87,678
Introducing 4D Primitive-Mâché (4DPM), a new method for replayable 4D reconstruction from monocular videos. We split dynamic scenes into 3D primitives and recover their motion. 4DPM can infer object positions even after they leave view. Joint work with @marwan_ptr @AjdDavison
5
25
179
33,370
more replies
4DPM handles multiple moving entities with distinct trajectories. The result is a complete, replayable 4D reconstruction of all observed scene elements at every timestamp.
1
1
5
876
Our method cuts 3D primitives out of the outputs of a feed-forward reconstruction model (π³) and then glues them across time. Each primitive’s motion is represented compactly as a single SE(3) pose, inferred from estimated 2D correspondences via an optimisation pipeline.
6
624
Load more