Shashank Tripathi
@sha2nk_t
Followers
1K
Following
1K
Media
43
Statuses
240
PhD student at Max Planck Institute for Intelligent Systems. Meta Research Fellow, 2023. prev: @amazon, @CarnegieMellon, @EpicGames, @Meta
Tübingen, Germany
Joined July 2013
Thanks #ICCV2023 for the amazing conference! I met many old friends and made scores of new ones. As a junior researcher, this opportunity to talk to thousands of people, with many of my heroes in the audience, was a humbling experience. I hope oral talk are there to stay.
2
2
37
📢 All of us will be at the PICO poster: Poster# 150, ExHall D from 10:30am to 12:30 pm today (13 June, Fri) at #CVPR2025. Come meet us there!.
If you’re not in freefall, you’re likely contacting something. Yet 3D human-object interaction (HOI) reconstruction remains underexplored. PICO (#CVPR2025) recovers humans 🏃♂️, objects 🏓, and their interactions 👉🍎- all in 3D, from just a single internet image. 1/11
0
0
6
@dimtzionas @akanazawa @Michael_J_Black @GerardPonsMoll1 @blacksquirrel__ @jhugestar And closing off with a panel discussion. If you have any burning questions.left unanswered, no better opportunity than this. Hosted by the wonderful @geopavlakos.
0
0
3
@dimtzionas @akanazawa @Michael_J_Black @GerardPonsMoll1 @blacksquirrel__ @jhugestar Our final talk in the series, Hanbyul Joo (@jhugestar) from Seoul National.University is speaking now on “Learning 3D Human-Object Interactions from.Generative Image Models”.
1
0
2
@dimtzionas @akanazawa @Michael_J_Black @GerardPonsMoll1 @blacksquirrel__ @jhugestar Now we have Sebastian Starke (@blacksquirrel__) from Meta, who will talk to us about “How to Break Your Neural Animation Controller”. Starting in 2 mins.
1
0
2
Relight images without rendering. Cool idea!.
🚀 Introducing GenLit – Reformulating Single-Image Relighting as Video Generation! We leverage video diffusion models to perform realistic near-field relighting from just a single image—No explicit 3D reconstruction or ray tracing required! No intermediate graphics buffers,
0
0
5
@dimtzionas @akanazawa @Michael_J_Black @GerardPonsMoll1 After a packed first session, we now break for a poster session and come back at 16:30 sharp for our.next speakers: Sebastian Starke (@blacksquirrel__) and Hanbyul Joo (@jhugestar). Posters are at boards #315-334, ExHall D.
1
0
1
@dimtzionas @akanazawa @Michael_J_Black Happening now: @GerardPonsMoll1 from University of Tubingen, speaking.on “How to Learn Large Scale 3D Human and Object/Scene Reconstruction Models."
1
0
0
@dimtzionas @akanazawa Next up, we have @Michael_J_Black from MPI, talking about “Humans in Motion and Contact: Towards Understanding Human Behavior in 3D”. Not to be missed.
1
1
2
@dimtzionas Our first talk is starting: @akanazawa from UC Berkeley on “How to Train Your Humanoids From Video”
1
0
3
🔔 Curious about the latest breakthroughs in 3D human understanding? Join us at the 2nd Workshop on 3D Human Understanding at #CVPR2025 for an exciting lineup of expert talks + panel discussion. Starts at 1:50 PM, Room 110b, June 12.
If you're at #CVPR2025, come by the Workshop on 3D Human Understanding tomorrow and meet all our amazing speakers in person!.🕑 June 12, 1:50 PM.📍 Room 110b.👤 @akanazawa @Michael_J_Black @GerardPonsMoll1 @blacksquirrel__ @jhugestar .🌐
0
1
5
Meet the winners and know how they did it at the #CVPR2025 Rhobin workshop:.🕑 9:00-12:00.📍Room 212.👉We also have an exciting lineup of speakers: @FedericaBogo, @ericyi0124, @angelayao101, @jiajunwu_cs, +panel discussion!.See you there.
🥁 📢🏅 The DAMON Challenge at the 3rd RHOBIN Workshop @ #CVPR2025 was a big success! With 51 teams and 108 submissions, it was fiercely competitive — pushing performance 38% to 122% beyond the DECO baseline in the Body and Semantic Contact Estimation tracks!!.
1
1
2
🥁 📢🏅 The DAMON Challenge at the 3rd RHOBIN Workshop @ #CVPR2025 was a big success! With 51 teams and 108 submissions, it was fiercely competitive — pushing performance 38% to 122% beyond the DECO baseline in the Body and Semantic Contact Estimation tracks!!.
🚀🚀 The 3rd RHOBIN challenge @CVPR on reconstruction of humans, objects, and interactions is live now! Past winners have gone on to publish in top conferences - it could be you this time! We also have exciting rewards from our sponsors. 🧵 1/3
0
2
14
RT @Michael_J_Black: One tweet. All the CVPR2025 projects that I’m part of. In one convenient place. Come see us in Nashville!. Conference….
0
26
0
PICO (#CVPR2025) is a great collaboration between @MPI_IS, @CarnegieMellon, @UTAustin, @UvA_Amsterdam, @meshcapade .Authors: @capicorn12, @sha2nk_t, @saidwivedi, Arjun Lakshmipathy, @AgnivC13, @Michael_J_Black, and @dimtzionas. Project Page:
0
0
7
PICO-fit handles 1-2 orders of magnitude more object instances than PHOSA and CONTHO or HDM. Shown here are HOI reconstructions from PICO-fit on new, previously untackled object categories such as bicycle, wineglass, hamburger, couch, etc. 10/11
1
1
3
PICO-fit’s contact predictions are more realistic and physically-plausible than those from CONTHO and PHOSA. Note that PICO-fit automatically recovers human pose, object geometry and scale, as well as their spatial alignment. 9/11
1
2
3
We evaluate PICO-fit on the out-of-domain Intercap dataset. PICO-fit significantly outperforms existing methods like HDM, CONTHO and PHOSA. PICO-fit’s reconstructions are also more realistic: humans prefer it over baselines 74.4% of the time. 8/11
1
1
3
(2) We use the 3D contact correspondences from PICO-db in a three-stage optimization framework PICO-fit. Stage 1 minimizes contact distance between human and object, stage 2 aligns 6DoF object pose to the image and stage 3 fine-tunes human limbs using image cues. 7/11
1
0
3