Learning and Optimization with 3D Orientations Alexandros Ntagkas, Constantinos Tsakonas, Chairi Kiourt, Konstantinos Chatzilygeroudis tl;dr: survey for orientation arxiv.org/abs/2509.17274

SLAM-Former: Putting SLAM into One Transformer Yijun Yuan, Zhuoguang Chen, Kenan Li, Weibang Wang, Hang Zhao tl;dr: e frontend and the backend promote each other with transformer arxiv.org/abs/2509.16909

Evict3R: Training-Free Token Eviction for Memory-Bounded Streaming Visual Geometry Transformers Soroush Mahdi, Fardin Ayar, Ehsan Javanmardi, Manabu Tsukada, Mahdi Javanmardi tl;dr: StreamVGGT+important token retention arxiv.org/abs/2509.17650

what makes a good paper video? what are some of the best you have seen? cc: @davidpicard.bsky.social @davidbau.bsky.social @eugenevinitsky.bsky.social @chriswolfvision.bsky.social @nsaphra.bsky.social @ducha-aiki.bsky.social @parskatt.bsky.social @martin-r-oswald.bsky.social @ericzzj.bsky.social

Wow, that's first time bsky paper tweet generates a similar amount of engagements to X. bsky.app/profile/duch...

Large Vision Models Can Solve Mental Rotation Problems Sebastian Ray Mason, Anders Gjølbye, Phillip Chavarria Højbjerg, Lenka Tětková, Lars Kai Hansen tl;dr: DINOv3,CLIP know 3D geometry, but only in middle layers. MAE bad, ImageNet bad, cls token bad. arxiv.org/abs/2509.15271

The CVPR 2026 deadline has finally been confirmed: Abstract: Nov 6th AoE Paper: Nov 13th AoE Location: Denver, USA cvpr.thecvf.com

VGGT vs MapAnything thread. Sorry for the other website link, but too many images to copy-paste. x.com/ducha_aiki/s...

MapAnything: Universal Feed-Forward Metric 3D Reconstruction Nikhil Keetha et 16 al. tl;dr: VGGT meets Pow3R. No RoPE. Metric scale. arxiv.org/abs/2509.13414 1/

Towards the Next Generation of 3D Reconstruction @parskatt.bsky.social PhD Thesis. tl;dr: would be useful in teaching image matching - nice explanations. (too) Fancy and stylish notation. Cool Ack section and cover image. liu.diva-portal.org/smash/record...

Last semester I have been asking ChatGPT to implement our computer vision course coding assignments. It is quite good there. More details in my blogpost. blog.dmytro.ai/posts/2025-0...

Why on Earth the local documentation for the TeX, which I never ever even opened, is the biggest part of the texlive????

And here is a link to the thesis itself: liu.diva-portal.org/smash/record...

Region-Wise Correspondence Prediction between Manga Line Art Images Yingxuan Li, Jiafeng Mao, Qianru Qiu, Yusuke Matsui tl;dr: train matching transformer for ViT patches -> handcrafted postprocess region & lines merging. arxiv.org/abs/2509.09501

LayerLock: Non-collapsing Representation Learning with Progressive Freezing Goker Erdogan, Nikhil Parthasarathy, Catalin Ionescu, Drew Hudson, Alexander Lerchner, Andrew Zisserman, Mehdi Sajjadi, Joao Carreira tl;dr: if your 1st layer has already converged -- freeze it. arxiv.org/abs/2509.10156

Faster VGGT with Block-Sparse Global Attention Chung-Shien Brian Wang, Christian Schmidt, Jens Piekenbrinck, Bastian Leibe tl;dr: block-sparse attention replaces global attention another work to improve scalability of VGGT arxiv.org/abs/2509.07120

If you have not idea why your docker build is failing, check your disk space, it is likely zero. Thank me later.

For some reason I am very happy for success of Kpop Demon Hunters and Silksong. Probably because both underpromised and overdelivered, and also respect users aka fans. And also being underdogs, ofc. Waiting for the some moment for the AI.

WHERE ARE THE GIFS?

WinT3R wins *3R naming contest hands down. arxiv.org/abs/2509.05296 github.com/LiZizun/WinT3R

🚀 new state-of-the-art on ILIAS dataset! Curious how well the latest models can recognize particular objects? We evaluated the base and large variants of DINOv3 and Perception Encoder (PE) on instance-level image retrieval. See the results 👉 vrg.fel.cvut.cz/ilias/

GeoArena: An Open Platform for Benchmarking Large Vision-language Models on WorldWide Image Geolocalization Pengyue Jia, Yingyi Zhang, Xiangyu Zhao, Yixuan Li tl;dr: in title arxiv.org/abs/2509.04334

From Editor to Dense Geometry Estimator JiYuan Wang et 9 al. tl;dr: repurposing editing DiT with VLM (Step1X-Edit) instead of plain DiT for depth/normals estimation + how to predict depth in bfloat16. arxiv.org/abs/2509.04338

The Telephone Game: Evaluating Semantic Drift in Unified Models Sabbir Mollah, Rohit Gupta Sirnam Swetha, Qingyang Liu  Ahnaf Munir, Mubarak Shah tl;dr: semantics gets lost quickly, BAGEL is the best. P.S. 2 years ago visual quality would be gone too arxiv.org/abs/2509.04438

Is there a virtual try-on paper/model, which is trained on real data(or sim), where the cloth can look bad on you because of too big/too small size? Not a rhetorical question.

I am 2 years late to the party, but @jessica.bsky.social essay on alignment aged very well. Also, I think that the most ifs here are already checked. joinreboot.org/p/alignment

I hope that the paper, which solves 3D reconstruction of from images with repeated patterns would have a subtitle “how it’s done, done, done”.

FastVGGT: Training-Free Acceleration of Visual Geometry Transformer You Shen, Zhipeng Zhang, Yansong Qu, Liujuan Cao tl;dr: token merging->VGGT without dense global attention arxiv.org/abs/2509.02560

Radially Distorted Homographies, Revisited Mårten Wadenbäck, Marcus Valtonen Örnhag, @parskatt.bsky.social tl;dr: closed form solution for homography meets radial distortion -> faster solver. arxiv.org/abs/2508.21190 P.S. "Unfortunately, HPatches does not have noticeable radial distortion" (c)

Here is input and output from nano-banana, GPT5, Flux-Kontext with the following prompt. Pretty useful, but not ready to make me a career in the game design :) "Make the animation sprite of this character, 4 frames per direction, move up, move left, move right, move down" P.S. Input (c)my daughter

Minimal Solvers for Full DoF Motion Estimation from Asynchronous Tracks @petrhruby.bsky.social, @marcpollefeys.bsky.social arxiv.org/abs/2508.17537

🧵 We're releasing CUS-QA - a new benchmark for testing LLMs on regional knowledge! Find out what your model knows about Czechia 🇨🇿, Slovakia 🇸🇰, and Ukraine 🇺🇦! 👉 Textual and visual questions, answers, and human judgment on model outputs! huggingface.co/datasets/ufa... www.arxiv.org/abs/2507.22752

HOSt3R: Keypoint-free Hand-Object 3D Reconstruction from RGB images Anilkumar Swamy, @vincentleroy.bsky.social @weinzaepfelp.bsky.social Jean-Sébastien Franco, Grégory Rogez tl;dr: DUSt3R tuned on hands-pose dataset + handcrafted 3D reconstruction system arxiv.org/abs/2508.16465

UnPose: Uncertainty-Guided Diffusion Priors for Zero-Shot Pose Estimation Zhaodong Jiang, Ashish Sinha, Tongtong Cao, Yuan Ren, Bingbing Liu, Binbin Xu tl;dr: diffusion for inpaint + MAST3R correspondences FTW arxiv.org/abs/2508.15972

ViPE: Video Pose Engine for 3D Geometric Perception Jiahui Huang et 14 al. tl;dr: SLAM-ish video processing: + BA + optical flow + metric monodepth arxiv.org/abs/2508.10934

Processing and acquisition traces in visual encoders: What does CLIP know about your camera? @ryan-ramos.bsky.social @stojnicv.xyz @gkordo.bsky.social Yuta Nakashima @gtolias.bsky.social @noagarciad.bsky.social tl;dr: CLIP sees difference DSLR vs iphone, DINO doesn't. arxiv.org/abs/2508.10637 1/

With Dino-v3 came new ViT sizes, but I think we are not done. I propose the following extensions: ViT-quantum ViT-footnotesize ViT-T ViT-S ViT-B ViT-L ViT-H ViT-g ViT-G ViT-7B ViT-Pro ViT-Pro Max ViT-Millenium ViT-Entreprise datacenter edition ViT-Supreme Leader ViT-Ultimate Showdown Turbo 3000

Having US-owned businesses in Ukraine will be a security guarantee in itself, they said. Reality check: 53% of US-owned businesses in Ukraine already have been targeted by russia and nothing happened. #UkrainianView #UAView united24media.com/war-in-ukrai...

🌟 Keynotes at #GCPR2025 🌟 🎤 Prof. Dima Damen (Uni Bristol & Google DeepMind) 🗓️ Thursday, Sept 25, 2025. 10:30–11:30 Talk:Opportunities in Egocentric Vision Discover new frontiers in egocentric video understanding, from wearable devices to large-scale datasets. 🔗 www.dagm-gcpr.de/year/2025/re...

Ultimate test for copilots in Cursor - extract the implementation of some object detector from mmdet abstraction hell.

🚀 DinoV3 just became the new go-to backbone for geoloc! It outperforms CLIP-like models (SigLip2, finetuned StreetCLIP)… and that’s shocking 🤯 Why? CLIP models have an innate advantage — they literally learn place names + images. DinoV3 doesn’t.

Have you ever asked yourself how much your favorite vision model knows about image capture parameters (e.g., the amount of JPEG compression, the camera model, etc.)? Furthermore, could these parameters influence its semantic recognition abilities?

More strikingly, we show that traces of these metadata labels (processing and acquisition parameters) can significantly impact the semantic recognition abilities.

DINOv3. Trained with hi-res refinement and GRAM-matrix supervision on gigantic dataset. Time to retrain all dinov2 based models :) Oriane Siméoni et 25 al. ai.meta.com/research/pub...

I have found ACing for NeurIPS this year a miserable experience and I've been struggling to articulate why without basically whining about it being too much work (although.....true. My god the amount of words I'm reading...). But I think I've figured out part of my dislike for this process.

Funny enough, it is slower to load VGGT weights from the disk, rather than run it for any amount of images, fitting into 24Gb video RAM

Training-free Geometric Image Editing on Diffusion Models Hanshen Zhu, Zhen Zhu, Kaile Zhang, Yiming Gong, Yuliang Liu, Xiang Bai arxiv.org/abs/2507.23300 tl;dr: handcrafted method (separately process background and object) + modification of attention and blending scheduling.

VLM-Guided Visual Place Recognition for Planet-Scale Geo-Localization Sania Waheed, Na Min An, Michael Milford , Sarvapali D. Ramchurn, Shoaib Ehsan tl;dr: in title arxiv.org/abs/2507.17455

Dens3R: A Foundation Model for 3D Geometry Prediction Xianze Fang et 10 al. tl;dr: hi-res, symmetric arch, "predict everything", version of DUST3R. No IMC eval. arxiv.org/abs/2507.16290

Registration beyond Points: General Affine Subspace Alignment via Geodesic Distance on Grassmann Manifold Jaeho Shin, Hyeonjae Gil, Junwoo Jang, Maani Ghaffari, Ayoung Kim tl;dr: optimization cost function for planes and lines registration. arxiv.org/abs/2507.17998