OutofAi (@OutofAi) header
OutofAi (@OutofAi) profile picture

OutofAi

@OutofAi

Simple ML.

 London, UK
github.com/OutofAi
Joined August 2023
  • Tweets443
  • Following155
  • Followers1.1K
95Media
Photo by OutofAi (@OutofAi)Photo by OutofAi (@OutofAi)Photo by OutofAi (@OutofAi)Photo by OutofAi (@OutofAi)Photo by OutofAi (@OutofAi)Photo by OutofAi (@OutofAi)Photo by OutofAi (@OutofAi)Photo by OutofAi (@OutofAi)Photo by OutofAi (@OutofAi)Photo by OutofAi (@OutofAi)
A Significantly simplified 2D Gaussian Splatting as part of a bigger research experiment regarding image signal processing that I am working on Github : github.com/OutofAi/2D-Gaussi… For those looking to gain a deeper understanding of Gaussian splatting, this is an excellent starting point. The observed noise during convergence is attributed to densification (pruning/cloning/splitting)
5
8
2
72
12,241
Thermal Infrared in Robotics.....................!! exciting! 🤔
Ayoung (@_ayoungk) profile pictureAyoung@_ayoungk6 Feb 2025
Exploring beyond RGB? Join us at ICRA 2025 for an exciting workshop on Thermal Infrared in Robotics (TIRO)! 🔥🤖 Share your latest thermal imaging projects in robotics or take part in our competition. 🔥 Workshop webpage sites.google.com/view/tiro25…
1
538
DeepSeek Is All You Need
1
2
467
Robotics needs its Gaussian Splatting moment
2
16
2,481
One of the most interesting paper that I read recently was arxiv.org/abs/2403.16990 Investigating two common issues with diffusion model image sampling, which the paper tackles by an training free approach, optimisation or guiding latents in the process of generation (1/3)
Be Yourself: Bounded Attention for Multi-Subject Text-to-Image Generation

Be Yourself: Bounded Attention for Multi-Subject Text-to-Image Generation

Text-to-image diffusion models have an unprecedented ability to generate diverse and high-quality images. However, they often struggle to faithfully capture the intended semantics of complex input...

arxiv.org
2
7
1,312
The first one being the cross-attention leakage, resulting similar subjects to shares traits that doesn't inherently belong to them (2/3)
Image from OutofAi (@OutofAi)
1
864
Thanks @TheAIAdvantage for the great review 🥲 you can watch the full clip on YouTube YT Link: youtu.be/7FCm_KZFJug?t=887 and if you wanna try Out of Focus v1.0: Github: github.com/OutofAi/OutofFocu… Colab: colab.research.google.com/gi…
4
27
8,515
When the paper claims that no additional training data is required and you get excited and then you realise you need 8xA100s to train the model💔
Image from OutofAi (@OutofAi)
1
6
657
Thanks for sharing our work🧡@_akhaliq
AK (@_akhaliq) profile pictureAK
@_akhaliq13 Sep 2024
Out of Focus v1.0 github: github.com/OutofAi/OutofFocu… Out of AI presents a flexible tool in Gradio to manipulate your images. This is our first version of Image modification tool through prompt manipulation by reconstruction through diffusion inversion process. We recommend running it on GPU with at least 22GB VRAM and preferabbly a A100 GPU for the best experience.
1
5
863
Out of Focus V1.0 Page: github.com/OutofAi/OutofFocu… We released our first version of our photo editing tool based on @Gradio. It attends to semantic similarities of prompts. By reconstructing given image in diffusion models. Whether you want the change the context of our product ad (🧵1/3)
5
29
192
22,407
or bring your 3d imagination to "life" (2/3)
1
2
14
10,666
We need the equivalent of Upwork but only AI agents being allowed. The model developers would get paid for tasks completed, but they are not allowed to interfere with the automated process! The agents are only allowed on the platform if they pass the eval for that category.
1
7
615
What AI robotics should focus on, instead of humanoids, an intelligent coffee table 😍😅😂
Giliam de Carpentier (@decarpentier_nl) profile pictureGiliam de Carpentier@decarpentier_nl9 Sep 2024
Enjoying the fruits of my labor now that I finished my 12-legged 'Carpentopod' table project. See decarpentier.nl/carpentopod for project info, or see it live at @EHVMMF this 13 & 14 Sept.
1
14
743
Our Cudacanvas module now hit 37,000 downloads, with 1k downloads per week, thanks everyone for using the package and your support ❤️ We probably will release a 3D version of the same package very soon to accommodate 3DGS in one notebook github.com/OutofAi/cudacanva…
GitHub - OutofAi/cudacanvas: Python Module for PyTorch Tensor Visualisation in CUDA Eliminating CPU...

GitHub - OutofAi/cudacanvas: Python Module for PyTorch Tensor Visualisation in CUDA Eliminating CPU...

Python Module for PyTorch Tensor Visualisation in CUDA Eliminating CPU Transfer - OutofAi/cudacanvas

github.com
8
608
With diffusion models tackling inverse problems for reconstruction and inferring underlying structure is the true test to evaluate how good a model can generalise!
7
721
We put together a simplified notebook based of Smoothed Energy Guidance (SEG) for Stable Diffusion 2.1 base and how this approach would significantly improve your results, along side CFG 🧵(1/3) Colab: colab.research.google.com/gi… Github: github.com/OutofAi/StableEne…
Image from OutofAi (@OutofAi)
Image from OutofAi (@OutofAi)
2
1
11
1,247
The approach is based of Guiding Diffusion Models with Reduced Energy Curvature of Attention paper, where the author points out that by reducing the underlying energy function, by applying Gaussian Blur on attention weights we can eliminate the negative effects of bad predictions 🧵(2/3)
1
1
170
Why complicate your repos when it can be just a Jupyter notebook file!🤦‍♀️
3
526
We created a series of simplified notebooks that cover essential aspects of Stable Diffusion, using the vanilla Stable Diffusion 2.1 base to utilise it as a face-editing model for building your own face app 🧵(1/3) Github : github.com/OutofAi/StableFac…
7
31
2
259
45,886
It covers Stable Diffusion topics regarding Sampling, CFG, architecture, Attention layer manipulation, DDIM inversion, pivotal tuning, real-world image reconstruction with SD and prompt2prompt (2/3)
Image from OutofAi (@OutofAi)
1
1
4
909
The rapid advancements in AI sometimes overshadow the untapped potential to utilise or monetise prior models, even with earlier iterations of diffusion architectures like Runway's Stable Diffusion 2.1 base.
3
526
One of the benefits you can get by installing and using our cudacanvas module is to visualise Stable Diffusion output in real-time, all on cuda with no cpu transfer (closing the window would also leads to early termination of sampling) > pip install cudacanvas
3
1
15
1,219
With little change to a normal Stable Diffusion pipeline and only the addition of a callback function.
Image from OutofAi (@OutofAi)
1
4
466
LLM companies must have users that just shuffle between them when new models come out! me included!
2
2
512
It seems a bit unintuitive to me that @GoogleColab still doesn't offer multiple T4 GPUs on a single instance whereas both Kaggle and PyTorch Lightning do in their notebooks ☹️☹️
1
8
1,256

No more items