Will Smith

Professor @UoY_CS, @YorkVGL group leader, vision/graphics/ML research, @Boro fan and climber (when 9 year old allows it) @willsmithvision.bsky.social

[Will Smith @WillSmithVision]

@daidailoh @sigbovik Perfect. Our paper wouldn't be complete without a precedence dispute over the idea!

Thanks for the nice comment. I'll email you the slides.

[Will Smith @WillSmithVision]

Excited to announce our latest (submitted to) SIGBOVIK 2026 @sigbovik paper: "SchmidhubAI: Accurate Historical Paper Attribution". We built an AI system that, given any modern AI paper, automatically determines which of its ideas were already published by Jürgen Schmidhuber.

[Will Smith @WillSmithVision]

@GaryMarcus @VisionBernie @SchmidhuberAI The entire premise is Schmidhuber was right about almost everything. The mean Schmidhuber Score for 10 landmark papers is 0.8970. We are, quite literally, agreeing with you. The joke isn't that he claims priority. It's that he's so consistently right you can automate the process.

[Will Smith @WillSmithVision]

To be clear: Schmidhuber genuinely has made enormous contributions to AI. The humour is affectionate. We fully expect him to respond with a thread explaining how he anticipated this paper in 1987.

[Will Smith @WillSmithVision]

The entire system is implemented as a CLAUDE.md file. Zero lines of executable code. Joint work with @VisionBernie. Paper: https://t.co/qoV24ovoAM

[Will Smith @WillSmithVision]

@AjdDavison @CSProfKGD Markdown

[Will Smith @WillSmithVision]

More details here: https://t.co/SpStV8n1ck

[Will Smith @WillSmithVision]

I am delighted (that pun will make sense in a second) that @AlistairFoggin's first paper, CroCoDiLight, has been accepted to ICLR. The idea came from a group discussion on the CroCo paper from @naverlabseurope and realising it might implicitly already understand relighting.

[Alistair Foggin @AlistairFoggin]

Very excited to announce my first paper, CroCoDiLight! Co-authored with @WillSmithVision and accepted at #ICLR2026.

Image relighting, shadow removal, and albedo estimation, all in our single base model with swappable task components.

See you in Rio @iclr_conf! [1/7]

[Will Smith @WillSmithVision]

The lighting representation is in image space so you can use it for things like lighting stabilisation or temporal upsampling in timelapse videos and lighting interpolation, but not currently for transferring lighting from one viewpoint to another - this is future work!

[Will Smith @WillSmithVision]

@jon_barron Be sure to mention that it's written in Fortran under the hood

[Will Smith @WillSmithVision]

@ali_shakiba_cs Easy! We already read the model weights checkpoint from a text file so no reason the dataset couldn’t also be loaded.