Lightnews — Scholar-powered news

Alexandre Morgand, PhD @alexmrgd.bsky.social · Aug 7

Ranran Huang, Krystian Mikolajczyk from Imperial College London

Project page: ranrhuang.github.io/spfsplat/
Paper: arxiv.org/abs/2508.01171
Source code (coming soon): github.com/ranrhuang/SP...

No Pose at All: Self-Supervised Pose-Free 3D Gaussian Splatting from Sparse Views

ranrhuang.github.io

Alexandre Morgand, PhD @alexmrgd.bsky.social · Aug 7

"No Pose at All Self-Supervised Pose-Free 3DGS from Sparse Views"
TLDR: 3DGS + no poses during training/inference; shared feature extraction backbone; simultaneous prediction of 3D Gaussian primitives+camera poses in a canonical space from unposed (1 feed-forward step).

1

Alexandre Morgand, PhD @alexmrgd.bsky.social · Jun 13

Yang Yang 1,2, Siming Zheng 2, Jinwei Chen 2, Boxi Wu 1, Xiaofei He 1, Deng Cai 1, Bo Li 2, Peng-Tao Jiang 2

1 Zhejiang University
2 VIVO MOBILE Communication Co., Ltd

Project page: vivocameraresearch.github.io/any2bokeh/
Paper: arxiv.org/abs/2505.21593
Source code: github.com/vivoCameraRe...

Any-to-Bokeh: One-Step Video Bokeh via Multi-Plane Image Guided Diffusion

a novel one-step video bokeh framework that converts arbitrary input videos into temporally coherent, depth-aware bokeh effects.

vivocameraresearch.github.io

Alexandre Morgand, PhD @alexmrgd.bsky.social · Jun 13

"Any-to-Bokeh: One-Step Video Bokeh via Multi-Plane Image Guided Diffusion"

📖TL;DR: Any-to-Bokeh is a novel one-step video bokeh framework that converts arbitrary input videos into temporally coherent, depth-aware bokeh effects.

1 1

Alexandre Morgand, PhD @alexmrgd.bsky.social · Jun 11

@sharathgirish97 1,2 @_TianyeLi 2* Amrita Mazumdar 2* @abhi2610 2 @davedotluebke 2 @shalinidemello 2

1 @umdglobalcampus
2 @nvidia

*Equal contributions

Alexandre Morgand, PhD @alexmrgd.bsky.social · Jun 11

Project page: research.nvidia.com/labs/amri/pr...
Paper: openreview.net/pdf?id=7xhwE...
Source code (released! NVIDIA license - non commercial): github.com/nvlabs/queen

QUEEN: QUantized Efficient ENcoding of Dynamic Gaussians for Streaming Free-viewpoint Videos

research.nvidia.com

1

Alexandre Morgand, PhD @alexmrgd.bsky.social · Jun 11

"QUEEN: QUantized Efficient ENcoding of Dynamic Gaussians for Streaming Free-viewpoint Videos"

TL;DR: Streamable free-viewpoint videos efficient representations for with dynamic Gaussians. Reduce model size to just 0.7 MB per frame while training in < 5s and rendering at 350 FPS

1 2

Alexandre Morgand, PhD @alexmrgd.bsky.social · May 20

Jiawei Yang *,¶, Jiahui Huang ¶, Yuxiao Chen ¶,
Yan Wang ¶, Boyi Li ¶, Yurong You ¶, Maximilian Igl ¶, Apoorva Sharma ¶, Peter Karkus ¶, Danfei Xu $,¶, Boris Ivanovic ¶, Yue Wang †,*,¶ Marco Pavone †,§,¶

* University of Southern California
$ GIT
§ Stanford University
¶ NVIDIA

† Equal advising

Alexandre Morgand, PhD @alexmrgd.bsky.social · May 20

Project page: jiawei-yang.github.io/STORM/
Paper: arxiv.org/abs/2501.00602
Source code: github.com/NVlabs/Gauss...

STORM

jiawei-yang.github.io

1

Alexandre Morgand, PhD @alexmrgd.bsky.social · May 20

"STORM: Spatio-Temporal Reconstruction Model for Large-Scale Outdoor Scenes"

TL;DR: Data driven transformer in a feed forward manner; dense reconstruction in dynamic environment with 3D gaussians and velocities; self-supervised scene flows

1 3

Alexandre Morgand, PhD @alexmrgd.bsky.social · Apr 22

@ucberkeleyofficial.bsky.social
, Max Planck Institute for Intelligent Systems, Stanford University

Project page: st4rtrack.github.io
Paper: st4rtrack.github.io/files/St4RTr...
Results: st4rtrack.github.io/page1.html

St4RTrack: Simultaneous 4D Reconstruction and Tracking in the World

st4rtrack.github.io

Alexandre Morgand, PhD @alexmrgd.bsky.social · Apr 22

(2/2) capturing static+dynamic scene geometry while maintaining 3D correspondences; long-range correspondences, effectively combining 3D reconstruction with 3D tracking; re-projection loss.

1 1

Alexandre Morgand, PhD @alexmrgd.bsky.social · Apr 22

St4RTrack: Simultaneous 4D Reconstruction and Tracking in the World

TL;DR: a feed-forward; (reconstructs+tracks dynamic video content); dust3r-like pointmaps for a pair of frames captured at different moments (1/2)

1 1 3

Alexandre Morgand, PhD @alexmrgd.bsky.social · Mar 14

Shangzhan Zhang 1,2*, @jianyuan_wang
3*, @YinghaoXu1
4*†, Nan Xue 2, Christian Rupprecht 3, @XiaoweiZhou5
1†, Yujun Shen 2, @GordonWetzstein 4

1 Zhejiang University
2 AntGroup
3 University of Oxford
4 Stanford University

*, † equal contributions (?)

Alexandre Morgand, PhD @alexmrgd.bsky.social · Mar 14

Project page: zhanghe3z.github.io/FLARE/
Paper: arxiv.org/pdf/2502.12138
Source code: github.com/ant-research...
Demo: huggingface.co/spaces/zhang...

FLARE

FLARE: Feed-forward Geometry, Appearance and Camera Estimation from Uncalibrated Sparse Views

zhanghe3z.github.io

1

Alexandre Morgand, PhD @alexmrgd.bsky.social · Mar 14

FLARE: Feed-forward Geometry, Appearance and Camera Estimation from Uncalibrated Sparse Views

TL;DR: feed-forward model; cascaded learning paradigm with camera pose serving as the critical bridge, recognizing its essential role in mapping 3D structures onto 2D image planes.

1 2

Alexandre Morgand, PhD @alexmrgd.bsky.social · Mar 13

Project page: fast3r-3d.github.io
Demo: fast3r.ngrok.app
Source code: github.com/facebookrese...

Fast3R: Towards 3D Reconstruction of 1000+ Images in One Forward Pass

Fast3R dramatically improves 3D reconstruction speed by processing up to 1500 images in a single forward pass.

fast3r-3d.github.io

Alexandre Morgand, PhD @alexmrgd.bsky.social · Mar 13

⚡️Fast3R: Towards 3D Reconstruction of 1000+ Images in One Forward Pass

TL;DR: multi-view generalization to DUSt3R; processing many views in parallel: Transformer-based architecture forwards N images in a single forward pass, bypassing the need for iterative alignment.

1 2

Alexandre Morgand, PhD @alexmrgd.bsky.social · Mar 12

Project page: ali-vilab.github.io/VACE-Page/
Paper: arxiv.org/pdf/2503.07598

VACE

All-in-One Video Creation and Editing

ali-vilab.github.io

1

Alexandre Morgand, PhD @alexmrgd.bsky.social · Mar 12

🪄 VACE: All-in-One Video Creation and Editing

from @alibabagroup.bsky.social's Tongyi Lab with:

Zeyinzi Jiang* Zhen Han* Chaojie Mao*† Jingfeng Zhang Yulin Pan Yu Liu

*Equal contribution, †Project lead

1

Alexandre Morgand, PhD @alexmrgd.bsky.social · Mar 10

From @nvidia (1), @NUSingapore (2), @UofT (3) and @VectorInst (4)

@jayzhangjiewu 1,2*, Yuxuan Zhang 1*, Haithem Turki 1, Xuanchi Ren 1,3,4, @JunGao33210520 1,3,4, Mike Zheng Shou 2, @FidlerSanja 1,3,4, @ZGojcic 1†, @HuanLing6 1,3,4†

*, † equal contribution

Alexandre Morgand, PhD @alexmrgd.bsky.social · Mar 10

Project page: research.nvidia.com/labs/toronto...
Paper: arxiv.org/abs/2503.01774

Difix3D+: Improving 3D Reconstructions with Single-Step Diffusion Models

research.nvidia.com

1

Alexandre Morgand, PhD @alexmrgd.bsky.social · Mar 10

Difix3D+: Improving 3D Reconstructions with Single-Step Diffusion Models

TL;DR: single-step diffusion models; a single-step image diffusion model trained to enhance and remove artifacts in rendered novel views caused by underconstrained regions of the 3D representation.

1 1

Alexandre Morgand, PhD @alexmrgd.bsky.social · Mar 5

Authors: Jovana Videnović , Alan Lukežič , Matej Kristan
from Faculty of Computer and Information Science, University of Ljubljana

Project page: jovanavidenovic.github.io/dam-4-sam/
Paper: arxiv.org/abs/2411.17576
Source code: github.com/jovanavideno...

DAM4SAM

Project page of the paper: A Distractor-Aware Memory for Visual Object Tracking with SAM2

jovanavidenovic.github.io

Alexandre Morgand, PhD @alexmrgd.bsky.social · Mar 5

A Distractor-Aware Memory (DAM) for Visual Object Tracking with SAM2

TL;DR: SAM2.1 based; distractor-distilled (DiDi) dataset to better study the distractor problem

1