Danila Rukhovich

ai research @ google

[Danila Rukhovich @filapro]

Z3D is accepted to #ACL2026!

multi-view images ➔ state-of-the-art 3D visual grounding

✅ no camera poses, no depths, no training
✅ outperforms prior SOTA by up to +40%
✅ combines Qwen3-VL, SAM3-Agent, CLIP, and MaskClustering

https://t.co/QLkphzkZUS
https://t.co/Qj69NA9MqN

[Danila Rukhovich @filapro]

𐂂𐂂𐂂 Our paper Zoo3D is accepted to #CVPR2026.

multi-view images ➔ state-of-the-art 3D detection

no camera poses, no depths, no training

https://t.co/0vbm0QkiBO
https://t.co/oDInekcxIj

[Danila Rukhovich @filapro]

cadrille is accepted to #ICLR2026
Qwen2 + SFT + RLVR ➡️ sota in 3D CAD generation

https://t.co/mRD6Q4DeA5
more details in 🧵⬇️

[Vladislav Kurenkov @vladkurenkov]

🚀 Introducing cadrille: a new SOTA model for CAD reconstruction from images, point clouds, and text—all in one framework with the use of RLVR.

Multimodal inputs + RLVR = clean, editable 3D models.

🧵👇

[Danila Rukhovich @filapro]

‼️Both our #3D #CAD papers are accepted at @ICCVConference #ICCV2025

CAD-Recode: Reverse Engineering CAD Code from Point Clouds
https://t.co/uMjz6rXvhN
🌟🌟🌟https://t.co/ZxvqC8iWR4

CAD-Assistant: Tool-Augmented VLLMs as Generic CAD Task Solvers
https://t.co/Ajo6V64KW5

[Danila Rukhovich @filapro]

🌎🌍🌏Check out our new remote sensing paper `MultiMAE Meets Earth Observation` accepted to #ICIP2025.

MAE pre-training on 6 modalities helps to achieve state-of-the-art results on 8 classification and segmentation benchmarks.

https://t.co/qxy6llqqMH
https://t.co/h7TOwok82o

[Danila Rukhovich @filapro]

@OpenMMLab 🔥UniDet3D is now accepted at #AAAI2025.

If interested in state-of-the-art indoor 3D object detection results, check our project page https://t.co/A8TITTjKbV. We release train / inference code based on @OpenMMLab MMDetection3D.

[Danila Rukhovich @filapro]

UniDet3D sets state-of-the-art on 6 indoor 3D object detection benchmarks with a single set of weights!

Weights and code based on @OpenMMLab mmdetection3d is released at https://t.co/EX2W5gHBYQ.

Full text is at https://t.co/lUgA1QP5Lf.

[Maksim Kolodiazhnyi @ma_ksi_mko]

🚀 Excited to share our new paper: UniDet3D: Multi-dataset Indoor 3D Object Detection! 🎉

We present a novel approach that enhances 3D object detection across multiple datasets, improving accuracy and versatility in indoor environments.
https://t.co/zs6ubHavop
#research #AI

[Danila Rukhovich @filapro]

@OpenMMLab 🔥 OneFormer3D is now accepted at @CVPR #CVPR2024 🔥

https://t.co/VrIiEvZQ3s <- for SOTA models in indoor 3D segmentation

[Danila Rukhovich @filapro]

Happy to announce OneFormer3D - our novel 3D segmentation framework that tackles semantic, instance, and panoptic segmentation tasks with a multi-task train-once design.

https://t.co/ubv7oxiMau
https://t.co/XvXzedzwVq

[Danila Rukhovich @filapro]

Finally, we published the code for OneFormer3D and pre-trained models (more to come soon). Thanks a lot @OpenMMLab for #mmdetection3d without which this work will not be possible.

https://t.co/VrIiEvZQ3s

[Danila Rukhovich @filapro]

@_akhaliq Also pleased to present our (with @ma_ksi_mko) novel 3D instance segmentation method TD3D. It demonstrates state-of-the-art results on ScanNet and S3DIS being faster on inference.

https://t.co/C5v65yaGCb
https://t.co/SIPr0UXESV

[Danila Rukhovich @filapro]

@_akhaliq Also pleased to present our (with @ma_ksi_mko) novel 3D instance segmentation method TD3D. It demonstrates state-of-the-art results on ScanNet and S3DIS being faster on inference.

https://t.co/C5v65yaGCb
https://t.co/SIPr0UXESV

[Danila Rukhovich @filapro]

@_akhaliq Our FCAF3D paper is finally accepted at #ECCV2022. Thanks @_akhaliq for sharing. And thanks @OpenMMLab for mmdetection3d framework without which this work would not be possible.

[AK @_akhaliq]

TR3D: Towards Real-Time Indoor 3D Object Detection

abs: https://t.co/LEmtghPteC
github: https://t.co/hbctC6UJ4V

[Danila Rukhovich @filapro]

@_akhaliq Our FCAF3D paper is finally accepted at #ECCV2022. Thanks @_akhaliq for sharing. And thanks @OpenMMLab for mmdetection3d framework without which this work would not be possible.