Can LLMs reason in fewer tokens?
At @_infocusp , we explored Token Assorted (ICML 2025). Couldn't find an existing implementation, so we built one!
Blog: infocusp.com/blogs/can-a-lan…
Code: github.com/infocusp/token_as…
Paper: openreview.net/forum?id=hYfO…
Code and Blog led by @JainilKotak
Hey @airtelindia, what a terrible broadband service you have! I am paying for 200mbps and this is what I get. Can't even listen to the speaker in an online meeting.
1
1
58
Milind Padalkar retweeted
My former institution @IITalk being presented by @alessiodelbue of @PAVIS_IIT. People definitely need the headphones with the noise around!
Milind Padalkar retweeted
We present 6DGS, our new @eccvconf paper on camera pose estimation using an NVS model! 📸🔬Our approach proposes a novel "radiant Ellicell" for 3DGS to quickly compare the image with the model.
Result? Faster, more accurate one-shot pose recovery.
🧵1/2
Can a humanoid become autonomous learning from teleoperation? Checkout this amazing work!
Here are the results of an exciting collaboration between PAVIS and the @ami_iit group at @IITalk. We deployed together a new human behaviour cloning model for humanoid robots!
Video: youtube.com/watch?v=zuFNEG62…
Paper: arxiv.org/abs/2406.15833
Project page: ami-iit.github.io/xbg/
1
57
Milind Padalkar retweeted
Join us now at @CVPR poster session (stand 379) and learn how we solve puzzles...and you can also challenge yourself with a real puzzle 🧩🧩🧩
Project: iit-pavis.github.io/DiffAsse…
Paper: arxiv.org/abs/2402.19302
#cvpr2024 @IITalk
And we bagged the third spot! 🥳🏆🥉
@PAVIS_IIT, @alessiodelbue, Dr. Severino Savarese, and everyone involved, thank you! – at Genoa, Liguria
Today @mgpadalkar is presenting his start-up project NirvanaVisor, using mixed reality and computer vision to support people with vision-related problems.
#startup #ai #mixedreality #computervision
@IITalk @alessiodelbue
Want to learn or practice #Yoga with an Indian teacher in #Genova? Get in touch with Dnyanada instagram.com/p/C4Q0p2aiiZk/… – at Genoa, Liguria
1
53
Need to solve a puzzle with randomly oriented pieces? Checkout this latest work from our lab, accepted in #CVPR2024!
🧩 Excited to unveil our latest #CVPR2024 paper! 🚀 DiffAssemble: a groundbreaking diffusion model solving puzzle reassembly challenges in both 2D and 3D spaces. Bye-bye noise, hello perfect reconstruction!
🎉 #GenAI #DiffusionModels
🔍 Dive into the details with our thread!👇
6
217
Predicting the next active object in first-person videos? Checkout what my labmate @sanketsans97 says at #SynapseAISymposium!
Replying to @PAVIS_IIT
and here @sanketsans97 presenting our work in collaboration with @vmurino, @BeyanCigdem, @pmorerio, @alessiodelbue
3
227
Milind Padalkar retweeted
How to avoid catastrophic forgetting in Source-Free UDA?
Take a look at "Continual Source-Free Unsupervised Domain Adaptation", Paper: arxiv.org/abs/2304.07374
by @waqar_ahmed89, @pmorerio and @vmurino
Presented today at @ICIAP2023 (oral) by @vmurino
Milind Padalkar retweeted
For more details, check our paper "VM-NeRF: Tackling sparsity in NeRF with View Morphing": arxiv.org/pdf/2210.04214
and our code here: github.com/mbortolon97/VM-Ne…
Milind Padalkar retweeted
Matteo Bortolon presenting today @ICIAP2023 a paper in collaboration with @Poiex @FBK_research and @alessiodelbue @IITalk. VM-NeRF improves NeRF when few cameras are available using view morphing.
Milind Padalkar retweeted
...always @ICIAP2023 , PAVIS has two posters this afternoon:
"Enhancing Hierarchical Vector Quantized Autoencoders for Image Synthesis Through Multiple" link.springer.com/chapter/10…
"Target-Driven One-Shot Unsupervised Domain Adaptation", arxiv.org/abs/2305.04628
🇮🇳🌖 What an historic event! So proud of @isro! #Chandrayaan3
Chandrayaan-3 Mission:
'India🇮🇳,
I reached my destination
and you too!'
: Chandrayaan-3
Chandrayaan-3 has successfully
soft-landed on the moon 🌖!.
Congratulations, India🇮🇳!
#Chandrayaan_3
#Ch3
1
119
Yet another interesting #ICCV2023 paper from our lab. Please have a look at it if you are interested in self-supervised 3D Keypoints extraction method.
Paper: arxiv.org/abs/2308.05410
Code: github.com/IIT-PAVIS/SC3K
Extracting reliable 3D keypoints from a point cloud is crucial for several tasks (e.g. object pose estimation).
Our new #ICCV2023 paper proposes a self-supervised method to extract coherent keypoints robust to rotated, decimated, noisy and intra-class perturbations.
How?
1/4
5
222
Milind Padalkar retweeted
L’app di MemEX raccontata da @alessiodelbue, responsabile Laboratorio Pavis #wnf22
Milind Padalkar retweeted
You know all about Ronaldo and Messi, now get the definitive story of the third highest scoring active men's international.
Sunil Chhetri | Captain Fantastic is available on FIFA+ now 🇮🇳
Milind Padalkar retweeted
𝐎𝐧𝐞 𝐟𝐨𝐫 𝐭𝐡𝐞 𝐡𝐢𝐬𝐭𝐨𝐫𝐲 𝐛𝐨𝐨𝐤𝐬! 🥈
Avinash Sable brings India's first-ever medal in men's steeplechase at Commonwealth Games
#B2022 | #IndiaAtB2022 | #AvinashSable | @avinash3000m
Milind Padalkar retweeted
Honourable mention for our paper in the MULA workshop @cvpr2020 "Semantically Grounded Visual Embeddings for Zero-Shot Learning" with @Shahnawazgrewal @I4c0p0 and @alessiodelbue




















