Shubham Goel

ML researcher. Previously, @berkeley_ai @iitbombay.

[Shubham Goel @goelshbhm]

@Guodzh Thanks for leading us, Guodong. Wish you all the best!!

[Shubham Goel @goelshbhm]

@imhaotian @elonmusk It been an honor working with you, Haotian. All the best, and enjoy Hawaii!

[Shubham Goel @goelshbhm]

@sumith1896 Congrats Sumith! 🔥🔥

[Shubham Goel @goelshbhm]

Friday 10:30AM Spotlight Poster#132

SAP3D is a system for 3D perception which improves as the number of views increases. Unposed few-view NVS/3D by test-time adaptation of diffusion NVS models.

[Yossi Gandelsman @YGandelsman]

[Poster alert - Friday morning #132]

Most of the 3D reconstruction and novel view synthesis systems take a fixed set of images as input. They can not improve when more unposed images are provided, even if they reveal new information about the 3D structure.

[Shubham Goel @goelshbhm]

Attending CVPR in Seattle, dm to meet and chat!

[Shubham Goel @goelshbhm]

@LeaMue27 @Michael_J_Black Congrats, Lea!

[Shubham Goel @goelshbhm]

@_tim_brooks @billpeeb @OpenAI Congrats guys, looks amazing!

[Shubham Goel @goelshbhm]

@nikhilaravi @Meta @hanna_mao @kirillov_a_n @tetexiao_ai Congrats!! Well deserved :)

[Shubham Goel @goelshbhm]

Thrilled to be presenting "Humans in 4D" at #ICCV.

We reconstruct humans in the wild with remarkable robustness and accuracy! That too, with a very simple transformerized architecture.

Thursday afternoon *Poster 76* in the South Foyer!

🌎 https://t.co/zvftvJoGj0

[Angjoo Kanazawa @akanazawa]

Humans in 4D, w/ @goelshbhm @geopavlakos @jathushan @JitendraMalikCV

✨HMR2.0💃🏻 + better PHALP, it's super robust! Give it a try:
🌎https://t.co/0YD6g5RfGA
🤗https://t.co/KiovKjE8WZ

*Thursday afternoon poster*

[Shubham Goel @goelshbhm]

@soalber00590884 @akanazawa @JitendraMalikCV @Avataar Our demo script (https://t.co/RhEOpM2XH0) for running HMR2.0 on an image folder supports saving meshes. For an example, see this cell of the collab: https://t.co/luBVMKEClT

[Shubham Goel @goelshbhm]

Happy to share that I’m PhD-done! Berkeley was such an enriching and cherishable experience, special thanks to my advisors @akanazawa and @JitendraMalikCV for supporting me throughout.

Next, super excited to join @Avataar, where we’ll continue pushing the boundaries of 3D tech.

[Shubham Goel @goelshbhm]

@soalber00590884 @akanazawa @JitendraMalikCV @Avataar Hi @soalber00590884 , thanks for your interest in our work. As mentioned in the GitHub readme, you can use the -- save_mesh flag with the demo script to generate .obj meshes.

[Shubham Goel @goelshbhm]

I’m attending SIGGRAPH in LA. Ping me if you’re also here and would like to meet!

[Shubham Goel @goelshbhm]

@geopavlakos @UTCompSci Congrats George!!

[Shubham Goel @goelshbhm]

Thanks @a16zGames for hosting an amazing event and inviting me to speak. I learnt so much!

[Troy Kirwin @tkexpress11]

Big questions discussed and debated:
- When will the Midjourney moment for 3D happen?
- How will AI NPC's transform gameplay?
- What will AI Native games look like?
- How should studio's think about copyright?
- What new gaming business models fall out from AI?

[Angjoo Kanazawa @akanazawa]

Releasing 🌟nerfstudio🌟, a plug-and-play python library to easily create your own NeRFs!
@nerfstudioteam is a contributor friendly open-source repo with a realtime web viewer that makes it easy to make cool videos 📽️
https://t.co/zw0ZPqIZ4c
https://t.co/NZM8GPjNrF #nerfstudio 1/

[Ben Poole @poolio]

Happy to announce DreamFusion, our new method for Text-to-3D!

https://t.co/4xI2VHcoQW

We optimize a NeRF from scratch using a pretrained text-to-image diffusion model. No 3D data needed!

Joint work w/ the incredible team of @BenMildenhall @ajayj_ @jon_barron

#dreamfusion

[Bill Peebles @billpeeb]

🎊 New paper!

We train loss-conditional diffusion models of *neural network checkpoints* that learn to optimize.

w/ Radosavovic, Brooks, Efros, Malik

proj: https://t.co/R58P8soLpy
arxiv: https://t.co/9AzqV0koIz
code: https://t.co/1nanm8v2WI

[Georgia Gkioxari @georgiagkioxari]

For recognition to graduate from 2D to 3D we need new tasks.

Omni3D is a large benchmark for 3D object detection in the wild with over 230k images & 3 million instances and draws from existing datasets.

Our model trained on Omni3D generalizes to new domains like AR captures

[Sasha Sax @iamsashasax]

Vision datasets (i.e. ImageNet) are usually collected once for a fixed task. But how do we know the choice of camera intrinsics, tasks, etc. is a good one?
Our ICCV paper on “steerable datasets” addresses this problem and gets 'human-level' surface normal preds along the way (⅓)