Yu-Ju Tsai

UC Merced EECS Ph.D. student

[Yu-Ju Tsai @Yu_Ju_Tsai]

Happy to share PROWBench 🚀
We explore how to systematically evaluate whether a world model can be programmed and controlled, rather than just generate plausible futures.
Check out the project below 👇

[Alaya Lab @alayastd]

Introducing PROWBench 🚀 — an extensible benchmark for Programmable World Models.

What does it mean for a world model to be programmable?
https://t.co/vvpVKUNhHO

[Yu-Ju Tsai @Yu_Ju_Tsai]

Happy to share EditHero, a project we’ve been working on! 🎉
We benchmark how well AI can edit 3D assets over multiple turns while precisely preserving what should stay unchanged.
More details below 👇

[Zhixiang Wang @lightchaserx]

Which AI is best at editing a 3D asset while preserving what you want to keep? 🥇🥈🥉

Introducing EditHero---a benchmark for long-horizon, part-level 3D editing & vibe modeling.

🌐 https://t.co/6cV7WICiJ6
🤗 https://t.co/lCIJTvw64s
💻 https://t.co/hQXNM4a98r

[Chieh Hubert Lin @ChiehHubertLin]

We are launching @app_illoca Tracing Paper today. illoca aims to bring an AI-first and AI-aided experience to architects. We are not just building fancy demos. We aim to provide real AI assistants that help eliminate the boring manual clicks and menu doom-scrolling. We want to integrate our AI into the actual workflow and help architects to materialize their creativity with minimal effort.

https://t.co/Yzm6rPjfKf

I am a believer that AI must be human-centric. Aesthetic taste, craftsmanship, and those final touches are what elevate a design into art. However, discussions about AI in design often carry a disheartening sentiment: the doomsday of creativity. AI research always grinds for novelty. The advancements are supposed to marry the best of two worlds (AI and design), and push creativity to the next level. Unfortunately, I observe the majority of the research remains focused on soulless AI mass production with limited controllability. This observation drives my PhD research and my work at Illoca. We put special focus on AI assistive tooling for the creative process, rather than making the decisions for the creator. Architects draft the vision, and our AI clears the friction.

Moreover, a primary pain point of AI in AEC is its data-driven nature. The lack of high-quality AEC data restricts most AI tools to merely rendering fancy images, rather than reading and understanding structural data. This is insufficient to resolve the sophisticated specifications and logistical requirements. Yet, this is exactly where the actual production bottleneck lies. Such a limited scope makes no real impact on the actual production. In AEC, the work is not just between a person and their AI assistants, but also includes the entire group of professionals from different industries behind the scenes. The need for collaboration shapes the industry to use structural and descriptive formats today, such as architectural drawings and CAD. At Illoca, we are building an AI system natively on these well-established representations; reading, thinking, and collaborating with you and your team.

This is just our first step, and there is more to come.

[Patrick Wu @tsunghan_wu]

All 5 --> Reject; All 4 --> Accept #NeurIPS2025

PC's Decision: Several reviewers, even those recommending acceptance, express dissatisfaction with the technical contributions.

The paper: Got 5 scores from ALL reviewers (5: Accept: Technically solid paper)

[Stan Szymanowicz @StanSzymanowicz]

⚡️ Introducing Bolt3D ⚡️
Bolt3D generates interactive 3D scenes in less than 7 seconds on a single GPU from one or more images.
It features a latent diffusion model that *directly* generates 3D Gaussians of seen and unseen regions, without any test time optimization.
🧵👇 (1/9)

[David Stutz @davidstutz92]

By popular request, I wrote a separate article on effectively managing and running experiments during a PhD in AI. I learned that effectively running experiments can be a crucial aspect of a PhD and I wanted to share some of my learnings 🧵: https://t.co/zDN4MAxlbC

[MrNeRF @janusch_patas]

OccFusion: Rendering Occluded Humans with Generative Diffusion Priors
https://t.co/g33ypiez9t
Project: https://t.co/I2ubwT7TaT
Method 1 | 2 ⬇️

[Yu-Ju Tsai @Yu_Ju_Tsai]

🔥 No More Ambiguity in 360° Room Layout via
Bi-Layout Estimation 🔥

Visit our #CVPR2024 poster to learn how we solve ambiguity challenge for 360° room layout estimation!

Project page: https://t.co/mzWGNv9YlV

🗓️ Friday, 06/21
🕔 17:00 - 18:30
📌 Poster 375

[Matthias Niessner @MattNiessner]

We will present Intrinsic Image Diffusion #CVP2024!

Material estimation is ambiguous -> we handle it with a probabilistic diffusion model. We obtain sharp albedo, roughness, and metallic maps just from a single image.

https://t.co/5ezM49e5no
By @Peter4AI with @vincesitzmann.

[An-Chieh Cheng @anjjei]

🌟Introducing "🤖SpatialRGPT: Grounded Spatial Reasoning in Vision Language Model" https://t.co/wxfUSvUw7s

SpatialRGPT is a powerful region-level VLM that can understand both 2D and 3D spatial arrangements. It can process any region proposal (e.g., boxes or masks) and provide answers to complex spatial reasoning questions.

🧵(1/n)

[Matthias Niessner @MattNiessner]

(1/6)
NeRF vs 3D Gaussians vs 3D Meshes

What's better? It's actually simple: the recent success in photo-realistic 3D reconstruction relies on efficient differentiable volumetric rendering. Rays across aligned images intersect, and we solve through ray integrals for the surface.

[Phillip Isola @phillip_isola]

Our computer vision textbook is released!

Foundations of Computer Vision
with Antonio Torralba and Bill Freeman
https://t.co/We0ZSJzkle

It’s been in the works for >10 years. Covers everything from linear filters and camera optics to diffusion models and radiance fields.

1/4

[#CVPR2026 @CVPR]

#CVPR2024 received 11,532 valid paper submissions, and only 2,719 were accepted, for an overall acceptance rate of about 23.6%

[Yu-Ju Tsai @Yu_Ju_Tsai]

The decision for #CVPR2024 is visible in the #Openreview!

[Alexey Bochkovskiy @alexeyab84]

YOLOv9 is better than any convolution or transformer based object detectors
Paper: https://t.co/0GmDLTjBRx
Code: https://t.co/4CMb04myyb

[Ziwei Liu @liuziwei7]

🔥Generative Gaussian Splatting🔥

#ICLR2024 Our 🚀DreamGaussian🚀 is accepted to @iclr_conf as **oral presentation**, enabling high-quality 3D generation in 2 minutes

- Project: https://t.co/VtVO63AKG7
- Code: https://t.co/edPso1vo8O
- Demo @huggingface: https://t.co/dRUroSxbl9

[AK @_akhaliq]

TikTok presents Depth Anything

Unleashing the Power of Large-Scale Unlabeled Data

paper page: https://t.co/fgga7kjUfb
demo: https://t.co/I7x17X4Qsi

Depth Anything is trained on 1.5M labeled images and 62M+ unlabeled images jointly, providing the most capable Monocular Depth Estimation (MDE) foundation models with the following features:

zero-shot relative depth estimation, better than MiDaS v3.1 (BEiTL-512)

zero-shot metric depth estimation, better than ZoeDepth

optimal in-domain fine-tuning and evaluation on NYUv2 and KITTI

[nerfstudio @nerfstudioteam]

🎉 V1.0.0 officially released!

* New splatfacto model, nerfstudio's implementation of Gaussian splatting
* Complete rewrite of the viewer, now built entirely on viser

[Jon Barron @jon_barron]

We just finished a joint code release for CamP (https://t.co/F5mmr9uHct) and Zip-NeRF (https://t.co/gdkqqzFpBC). As far as I know, this code is SOTA in terms of image quality (but not speed) among all the radiance field techniques out there. Have fun! https://t.co/xb5irSyHY0

[Yu-Ju Tsai @Yu_Ju_Tsai]

🎊Happy to announce that our paper 'Dual Associated Encoder for Face Restoration' has been accepted to #ICLR2024 @iclr_conf!
Thank co-authors for all the valuable advice!
See you in Vienna!

Paper: https://t.co/wWutd3i13D
Project page: https://t.co/1aOZ0QWwKO