@marwan_ptr

PhD student at the Dyson Robotics Lab @ Imperial College London.

marwan99.github.io/
Joined April 2011
  • Tweets79
  • Following509
  • Followers504
10Media
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
6
742
87,717
Simify: use generative real2sim to build a simulator, then perform spatial reasoning using evolutionary search. Presented by @YafeiHuCMU at IROS! arxiv.org/abs/2609.33982

Test-Time Spatial Reasoning for Robot Manipulation Using...

Spatial reasoning is fundamental to general robot intelligence, as it enables robots to complete long-horizon tasks involving multi-object interaction. We introduce Simify, a training-free,...

arxiv.org
1
3
2
25
3,863
Marwan Taher retweeted
Kirill Mazur@makezurSep 14
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
9
430
79,188
Download the code and give it a try!
Andrew Davison
@AjdDavisonJun 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.
1
4
12
2,354
Project page: marwan99.github.io/kv_tracke… Code: github.com/Marwan99/kv_track…
2
113
Wanna see a real-time demo of KV-Tracker? Stop by in ExHall F today 11:45 - 13:45! #CVPR2026
2
12
1
54
4,964
Project page: marwan99.github.io/kv_tracke… Code: github.com/Marwan99/kv_track…

KV-Tracker: Real-Time Pose Tracking with Transformers

Real-time 6-DoF pose tracking by caching key-value pairs from multi-view geometry networks. Up to 15× speedup at ~30 FPS without drift or catastrophic forgetting.

marwan99.github.io
1
1
230
Marwan Taher retweeted
Kirill Mazur@makezurJun 5
How to build complete 4D reconstructions from videos? Swing by 4DPM oral presentation today: Mile High Ballroom 3A - 4A, 13:00 #CVPR2026
4
8
2
108
11,032
Marwan Taher retweeted
Kirill Mazur@makezurApr 16
4DPM got selected for an oral presentation at #CVPR2026 🤠
Kirill Mazur@makezur18 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
7,210
KV-Tracker has been accepted to #CVPR2026!
Marwan Taher@marwan_ptr19 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
19
223
16,928
KV-Tracker enables object-level reconstruction and tracking when provided with an object mask. The KV-cache can be saved and used later without any special initialisation procedure.
2
19
2,366
More results and the paper can be found here: marwan99.github.io/kv_tracke… Video: youtu.be/ZVNnvZZxhoI

KV-Tracker: Real-Time Pose Tracking with Transformers

Real-time 6-DoF pose tracking by caching key-value pairs from multi-view geometry networks. Up to 15× speedup at ~30 FPS without drift or catastrophic forgetting.

marwan99.github.io
1
32
3,447
Per-frame geometry from π³ is split into primitives via segmentation and tracked over time using dense 2D point tracks. With a compact per primitive pose, geometry is densely aligned, stitching primitives to create a complete reconstruction of the observed scene components.
Kirill Mazur@makezur18 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
9
584
ACE-SLAM naturally handles loop closure without special treatment, robustly deals with dynamic objects, while remaining lightweight (small MLP) and computationally efficient—making this representation compelling for SLAM.
Ignacio Alzugaray@alzugarayign17 Dec 2025
Excited to present ACE-SLAM, the first neural SLAM to use Scene Coordinate Regression as an implicit map representation Efficient (real-time from live stream), compressive (neural maps <1MB) and robust to dynamic scenes With @marwan_ptr and @AjdDavison ialzugaray.github.io/ace-sla…
2
10
1,006
Marwan Taher retweeted
Xin Kong@XinKong_IC9 Sep 2025
🚀 Excited to share CausNVS: Autoregressive Multi-view Diffusion for Flexible 3D Novel View Synthesis! Let’s recon 3D world generatively. CausNVS handles any number of input views, synthesizes novel views autoregressively, enables interactive streaming and flexible N-to-M NVS.
2
18
107
10,339
Marwan Taher retweeted
Riku Murai@rmurai061016 Dec 2024
Introducing MASt3R-SLAM, the first real-time monocular dense SLAM with MASt3R as a foundation. Easy to use like DUSt3R/MASt3R, from an uncalibrated RGB video it recovers accurate, globally consistent poses & a dense map. With @eric_dexheimer*, @AjdDavison (*Equal Contribution)
42
251
38
1,424
204,036
EscherNet will be presented tomorrow at #CVPR. But *now* you can drop a couple of images into our Hugging Face demo to try it out! huggingface.co/spaces/kxic/E…

EscherNet - a Hugging Face Space by kxic

3D novel view synthesis from any number images!

huggingface.co
Xin Kong@XinKong_IC19 Jun 2024
Tired of single image to 3D? Check out EscherNet tomorrow @CVPR that can take flexible number of views for 3D generation! THURSDAY, JUNE 20 ORAL: 9:00-10:30, SUMMIT BALLROOM (TOP FLOOR) POSTER: 10:30-12:00, ARCH 4A-E, #69 Try our @Gradio online demo huggingface.co/spaces/kxic/E…
6
434
Load more