Sabitlenmiş Tweet
Introducing Vision Banana🍌: an image generator that achieves SOTA on segmentation, depth prediction, and surface normal estimation 🚀
🖼️Project page: vision-banana.github.io
📜Technical report: arxiv.org/abs/2604.20329
🧵👇
English
Valentin Gabeur
101 posts
Valentin Gabeur
@vgabeur
Research Scientist @GoogleDeepmind | Prev. Postdoc @MetaAI, PhD @Inria & @GoogleAI
San Francisco Katılım Mart 2012
338 Takip Edilen614 Takipçiler
@alcinos26 @sainingxie @jalayrac @jon_barron Thanks for the feedback Nico, I listed justifications for our decisions in the following thread:
x.com/vgabeur/status…
Valentin Gabeur@vgabeur
Here is a thread with some precisions regarding our evaluation of Vision Banana on the SA-Co/gold segmentation benchmark. [1/n] 🧵👇
English
In this age of PR, it's common to see bombastic claims like "beating SAM3". However I take issue with this chart which is quite dishonest IMHO.
I would have expected more academic honesty from researchers I deeply respect @sainingxie, @vgabeur, @jalayrac @jon_barron.
A quick 🧵
Saining Xie@sainingxie
the idea of (using image generators to solve perception tasks) is pretty straightforward, and there have been many interesting results over the past couple of years. so why this moment matters? because for the first time, a single generalist model is actually beating top domain-specific models like SAM3 and DepthAnything3. those specialized models usually take years to develop and rely on pretty complex recipes in training and data. yet, as history often shows, such capabilities can instead emerge from general, scalable pretraining. in this case, image editing turns out to be a really effective pretraining paradigm, and all of the dense labeling problems can just be reframed as post-training on top of that. [2/n]
English
All in all, I agree that our evaluation on SA-Co/gold is incomplete with respect to the official setup recommended in the SAM3 paper, but we still wanted to share it, to show that instance segmentation is challenging and our model is not quite there yet.
[9/n]
English
Our classification of SAM3 as "non zero-shot" on SA-Co/gold was not only guided by 93% of the NPs being seen during SAM3 training, but also by the annotation process similarities between SA-Co/gold and SAM3 training data (e.g. image collection, mask annotation tool).
[8/n]
English
Here is a thread with some precisions regarding our evaluation of Vision Banana on the SA-Co/gold segmentation benchmark.
[1/n]
🧵👇
Nicolas Carion@alcinos26
In this age of PR, it's common to see bombastic claims like "beating SAM3". However I take issue with this chart which is quite dishonest IMHO. I would have expected more academic honesty from researchers I deeply respect @sainingxie, @vgabeur, @jalayrac @jon_barron. A quick 🧵
English
Huge thanks to my teammates!
@ShangbangLong, @songyoupeng, @PaulVoigtlaend1, @Kevin_SSY, Yanan, Karen, Zhicheng, Wenlei, @jon_barron, Kyle, Nithish, Sherry, Yandong, Mandy, Suhas, Yiming, Huizhong, @oliver_wang2, @sainingxie, @howardzzh, Kaiming, Tom, @jalayrac, @RSoricut
Filipino
Image generation is becoming the universal interface for computer vision, just as text generation did for language and reasoning. Generative vision pretraining is paving the way for true Foundational Vision Models.
English
Introducing Vision Banana🍌: an image generator that achieves SOTA on segmentation, depth prediction, and surface normal estimation 🚀
🖼️Project page: vision-banana.github.io
📜Technical report: arxiv.org/abs/2604.20329
🧵👇
English
Valentin Gabeur retweetledi
We’re launching Nano Banana 2, built on the latest Gemini Flash model. 🍌
It’s state-of-the-art for creating and editing images, combining Pro-level capabilities with lightning-fast speed. 🧵
GIF
English
Valentin Gabeur retweetledi
An updated Gemini 3 Deep Think is out today:
📈 Achieves SOTA on ARC-AGI-2, MMMU-Pro, and HLE.
🥇Gold-medal level on Physics & Chemistry Olympiads.
It turns out the best way to solve hard problems is still to think about them. Read more: bit.ly/4kzBLqq
English
Valentin Gabeur retweetledi
Introducing Agentic Vision — a new frontier AI capability in Gemini 3 Flash that converts image understanding from a static act into an agentic process.
By combining visual reasoning with code execution, one of the first tools supported by Agentic Vision, the model grounds answers in visual evidence and delivers a consistent 5-10% quality boost across most vision benchmarks. Here’s how the agentic ‘Think, Act, Observe’ loop works:
— Think: The model analyzes an image query then architects a multi-step plan
— Act: The model then generates and executes Python code to actively manipulate or analyze images
— Observe: The transformed image is appended to the model's context window, allowing it to inspect the new data before generating a final response to the initial image query
Learn more about Agentic Vision and how to access it in our blog ⬇️
blog.google/innovation-and…
English
Valentin Gabeur retweetledi
