Ahan

41 posts

Ahan banner
Ahan

Ahan

@ahan_sh

PhD student @SFU

Katılım Kasım 2013
597 Takip Edilen247 Takipçiler
Sabitlenmiş Tweet
Ahan
Ahan@ahan_sh·
Excited to share our recent work: Free-Range Gaussians 🥚✨ The core idea: instead of predicting Gaussians on a pixel- or voxel-aligned grid, we let them live freely in 3D space. 🌐 Project: free-range-gaussians.github.io 📝 Paper: arxiv.org/abs/2604.04874
English
14
86
606
73.3K
Ahan retweetledi
Andrea Tagliasacchi 🇨🇦
Andrea Tagliasacchi 🇨🇦@taiyasaki·
📢📢📢introducing 𝐏𝐨𝐰𝐞𝐫 𝐅𝐨𝐚𝐦 A 3D representation that can be ray traced or rasterized in real time, with NO COMPROMISE in quality. - Project: powerfoam.github.io - arXiv: arxiv.org/abs/2604.24994 Rasterized at 3DGS-class FPS Ray traced at Radiant Foam speeds
English
19
111
744
91.3K
Ahan
Ahan@ahan_sh·
@NicholasBardy Yes, in pure GS optimization, Gaussians are already coordinate-based. Our point is that many predictive/generative methods introduce pixel-/voxel-aligned output structure. We bring back the original free 3D parameterization to the feed-forward/generative setting
English
1
0
0
31
Nicholas Bardy
Nicholas Bardy@NicholasBardy·
@ahan_sh I'm confused what "grid-aligned" output is. Haven't Gaussians always been coordinated based, not pixel based? What am I missing here?
English
1
0
0
21
Ahan
Ahan@ahan_sh·
Excited to share our recent work: Free-Range Gaussians 🥚✨ The core idea: instead of predicting Gaussians on a pixel- or voxel-aligned grid, we let them live freely in 3D space. 🌐 Project: free-range-gaussians.github.io 📝 Paper: arxiv.org/abs/2604.04874
English
14
86
606
73.3K
Ahan
Ahan@ahan_sh·
@thirdachilles Related, but not quite. Differentiable rendering is one ingredient here, but the main contribution is modeling a scene as a set 3D Gaussian splats that are directly modeled with a transformer.
English
0
0
4
819
Timur Fattahov
Timur Fattahov@thirdachilles·
@ahan_sh Is this like differentiable rendering but with splats?
English
1
0
0
996
Ahan
Ahan@ahan_sh·
To scale to larger Gaussian sets, we use hierarchical patching / a level-of-detail tree to group spatially related Gaussians into joint tokens (similar to patching in image diffusion)
Ahan tweet media
English
1
0
9
1.6K
Ahan retweetledi
Jan Held
Jan Held@janheld14·
🚀 I’m excited to share my final work as a PhD student: 𝙈𝙚𝙨𝙝𝙎𝙥𝙡𝙖𝙩𝙩𝙞𝙣𝙜: 𝘿𝙞𝙛𝙛𝙚𝙧𝙚𝙣𝙩𝙞𝙖𝙗𝙡𝙚 𝙍𝙚𝙣𝙙𝙚𝙧𝙞𝙣𝙜 𝙬𝙞𝙩𝙝 𝙊𝙥𝙖𝙦𝙪𝙚 𝙈𝙚𝙨𝙝𝙚𝙨 - Arxiv: arxiv.org/abs/2512.06818 - Code: github.com/meshsplatting/… - Project page: meshsplatting.github.io
English
15
138
884
78.4K
Ahan retweetledi
Arip
Arip@machinestein·
While we are going back to the era of research… Introducing 𝗗𝗲𝗲𝗽 𝗜𝗺𝗽𝗿𝗼𝘃𝗲𝗺𝗲𝗻𝘁 𝗦𝘂𝗽𝗲𝗿𝘃𝗶𝘀𝗶𝗼𝗻 (𝗗𝗜𝗦) – a new learning method for recursive reasoning. DIS builds on the elegant Tiny Recursive Model (TRM)(@jm_alexia) but makes recursion radically simpler: - 𝟏𝟖× 𝗳𝗲𝘄𝗲𝗿 𝗳𝗼𝗿𝘄𝗮𝗿𝗱 𝗽𝗮𝘀𝘀𝗲𝘀 - 𝗡𝗼 𝗵𝗮𝗹𝘁𝗶𝗻𝗴 𝗺𝗲𝗰𝗵𝗮𝗻𝗶𝘀𝗺 - And a tiny 0.8M-parameter model reaching 24% accuracy on ARC-AGI-1 (@arcprize) Paper: arxiv.org/pdf/2511.16886 Code: github.com/machinestein/D…
Arip tweet media
English
1
6
22
5.3K
Ahan retweetledi
Sauradip Nag
Sauradip Nag@Dumb_Thug·
Thanks for sharing @janusch_patas ! TL;DR - Sparse Views to Dynamic 3D generation. Key Insight - Slowing a video (magnifying temporally) improves 3D generation quality.
MrNeRF@janusch_patas

In-2-4D: Inbetweening from Two Single-View Images to 4D Generation Contributions: • To the best of our knowledge, In-2-4D is the first method for generative 4D inbetweening over two distant monocular frames spanning arbitrary motions. • Our novel hierarchical approach breaks the complex inbetweening into a series of simpler motion estimations via video, followed by 4D (i.e., dynamic 3DGS) generation. • To generate smooth 3D object and motion transitions, we further optimize the 3D trajectories using a bottom-up merging strategy with smoothing regularization. • We contribute a new 4D interpolation benchmark, I4D-15, on challenging object motions and real-world scenes.

English
1
3
16
2.5K
Ahan retweetledi
Miguel Angel Bautista
Miguel Angel Bautista@itsbautistam·
Here's one to read on your flight to #NeurIPS2024! A flow-matching transformer model in function space! This model has all the advantages of neural fields: resolution-free generation and domain-agnostic architecture, while obtaining strong results on ImageNet-256 and Objaverse!
Yuyang Wang@YuyangW95

1/n 🚨New preprint! Our work “Coordinate In and Value Out: Training Flow Transformers in Ambient Space” arxiv.org/abs/2412.03791 presents a domain-agnostic and end2end flow-matching generative model that effectively handles various modalities like images and point clouds.

English
2
12
63
6.8K
Ahan retweetledi
Andrea Tagliasacchi 🇨🇦
Andrea Tagliasacchi 🇨🇦@taiyasaki·
📢📢📢 RoMo: Robust Motion Segmentation Improves Structure from Motion romosfm.github.io arxiv.org/pdf/2411.18650 TL;DR: boost your SfM pipeline on dynamic scenes. We use epipolar cues + SAMv2 features to find robust masks for moving objects in a zero-shot manner. 🧵👇
English
2
13
98
18.6K
Ahan retweetledi
Sherwin Bahmani
Sherwin Bahmani@sherwinbahmani·
📢 Excited to share our new work: AC3D: Analyzing and Improving 3D Camera Control in Video Diffusion Transformers snap-research.github.io/ac3d We analyze what pre-trained video diffusion transformers understand about 3D and demonstrate dynamic scene generation with 3D control.
English
6
23
119
15.7K
Ahan retweetledi
Ilya Petrov
Ilya Petrov@ptrvilya·
Blendify 2.0 is out!🎉 This major release of our Python framework for creating and rendering scenes in Blender introduces several new features. More details below 🧵 (1/3). GitHub: github.com/ptrvilya/blend… Developed together with @guzov_vladimir.
English
1
3
15
662

Keşfet

@NicholasBardy @thirdachilles @taiyasaki @PeterHedman3 @ethanjohnweber @c_richardt @DRozumnyi @jm_alexia