iiinova — where your digital self keeps evolving

Christian Wolf @chriswolfvision.bsky.social · 08/10/2026

OMG this saves so much time. I have my own Jarvis now.

Christian Wolf @chriswolfvision.bsky.social · 08/10/2026

Me: writing the paper on overleaf Claude: sidekick, ready to do reformatting, layout, eg. remove white space in front of each paragraph; integrate figure; use tikz to overlay text.

250

Christian Wolf @chriswolfvision.bsky.social · 07/10/2026

Recurrent Looped Transformer Zhang et al. Combine context/history transformer with recurrence: have a causal encoder with a recurrent decoder. arxiv.org/abs/2610.07591

1171

Christian Wolf @chriswolfvision.bsky.social · 07/10/2026

I am registered for the Paris venue of NeurIPS 2026! See you there 😀

2120

Christian Wolf @chriswolfvision.bsky.social · 06/10/2026

Procedural Core: A Compact Recurrent Initialization for Vision Transformers By @damienteney.bsky.social 's group Idea: train a ViT with parameters tied over layers (="recurrent ViT") on abstract procedurally data. Then initialise models of ANY arch (!) with these weights. arxiv.org/abs/2609.37631

061

Christian Wolf @chriswolfvision.bsky.social · 06/10/2026

Having fun with Blender ... Distilling context transformers into recurrent transformers: arxiv.org/abs/2606.21562 @weinzaepfelp.bsky.social @mbsariyildiz.bsky.social

1225

Christian Wolf @chriswolfvision.bsky.social · 01/10/2026

Je sens que Bluesky est pret pour des memes sur la causalité.

5235

Christian Wolf @chriswolfvision.bsky.social · 25/09/2026

In the meantime, back on the hellsite 😬

020

Christian Wolf @chriswolfvision.bsky.social · 23/09/2026

Rookie mistake: using the same conda environment in different projects which are worked on by different Claude instances. That was not smart.

030

Christian Wolf @chriswolfvision.bsky.social · 23/09/2026

Claude when I now ask it to re-check (with Opus 5.5) the codebase already checked by Fable 5.1 and initially implemented by Opus 4.8.

060

Christian Wolf @chriswolfvision.bsky.social · 23/09/2026

The appropriate measure when somebody asks for too many CPUs on a single GPU job and the other GPUs of that node are now idling.

061

Christian Wolf @chriswolfvision.bsky.social · 20/09/2026

Let's make this a thread ... Lyon is seen in the background, with the towers at La Part Dieu. The hill I am standing on is Mt. Thou. Mt. Cindre is in the background. The highest elevation is unfortunately only 629m. This is not Grenoble 😬

060

Christian Wolf @chriswolfvision.bsky.social · 17/09/2026

My 10,000th kilometerversary! Celebrating 10,000 e-bike kms 🚲commuting between Lyon and Grenoble (with some help from French TER trains which are frequently delayed).

3230

Christian Wolf @chriswolfvision.bsky.social · 16/09/2026

Red is the current model. All other colors are proposed "improvements". Higher is better. I should stop proposing improvements I guess.

380

Christian Wolf @chriswolfvision.bsky.social · 16/09/2026

"Maniac Mansion" from Lucas Arts. It started this type of games, upending the run of Sierra, and then much later led to "Monkey Island 2" on PC.

120

Christian Wolf @chriswolfvision.bsky.social · 15/09/2026

One more reason to work on real robotics with real robots: your papers including experiments taken humans days and weeks will stand out compared to the floods of papers which took Claude or Codex a couple of seconds to generate. If somebody finds them, that is. (Graph taken from the quoted thread)

1100

Christian Wolf @chriswolfvision.bsky.social · 13/09/2026

C'est tellement plus logique écrit comme ça...

060

Christian Wolf @chriswolfvision.bsky.social · 12/09/2026

Après-conference with Diane and Philippe.

1140

Christian Wolf @chriswolfvision.bsky.social · 12/09/2026

Evaluated on 966 real navigation episodes: real robot, real building arxiv.org/abs/2606.21216

000

Christian Wolf @chriswolfvision.bsky.social · 12/09/2026

ECCV Poster #305 at 3pm (now): A scalar per patch from pre-trained ViTs enables fast moving navigation in the real world @steevenj7.bsky.social, L. Antsfeld, myself (I will present) - Visual enc distilled from heterog. teachers - Attention proj - Pretrain Policy w Lidar input => remove Lidar

1154

Christian Wolf @chriswolfvision.bsky.social · 12/09/2026

ECCV Panel on 'Can AI build new knowledge' with @dimadamen.bsky.social @yilundu.bsky.social and Viorica Patraucean.

051

Christian Wolf @chriswolfvision.bsky.social · 12/09/2026

Jamie Shotton's keynote: autonomous driving with world models at Wayve.

181

Christian Wolf @chriswolfvision.bsky.social · 12/09/2026

Jamie Shotton's keynote at ECCV on what at that time completely changed my mind on Microsoft, and won CVPR 2011 best paper, CVPR 2021 test of time awards. The Primesense Kinect depth sensor used in this project revolutionised robotics.

0102

Christian Wolf @chriswolfvision.bsky.social · 11/09/2026

At the ECCV 2026 reception 🕺👯💃

080

Christian Wolf @chriswolfvision.bsky.social · 10/09/2026

FEEL (Force-Enhanced Egocentric Learning): A Dataset for Physical Action Understanding Dessalene et al., UMD #ECCV2026 oral arxiv.org/abs/2603.15847 Very nice!

1162