Sivan Doveh
@SivanDoveh
Followers
242
Following
535
Media
34
Statuses
237
Context-Aware VLMs; CS Ph.D. Candidate @ Weizmann; Student Researcher @Google
Joined January 2018
RT @ramoscsv: Deadline Extended!!!. Submit your paper to the LongVid-Foundations Workshop @ICCVConference and make part of the discussion!โฆ.
0
5
0
RT @ramoscsv: ๐ Last 24 hours!! ๐. Donโt shelve that great idea! .Submit your paper to the LongVid-Foundations Workshop @ICCVConference andโฆ.
0
4
0
@jmie_mirza @NimrodShabtay @WeiLinCV @ArbelleAssaf @Eli_Schwartz @RogerioFeris @HildeKuehne James Glass @leonidkarlinsky Shimon Ullman.
0
0
7
IPLOC accepted to ICCV25 โบ๏ธ.Thanks to all the people that were part of it ๐ฉท. The idea for this paper came by a lake during a visit to Graz for a talk. It has traveled with me through too many countries and too many wars, and itโs now a complete piece of work.
Ever wanted to locate your cat in a database of images using just one reference image? Probably notโbut this highlights a gap in VLMs. They struggle to localize specific objects given in-context examples, often copying the last sample's location instead of learning from it.
2
2
15
Working on videos that are longer than 8 seconds? Want to visit Hawaii? Consider submitting to this workshop ๐. LongVid-Foundations @ICCVConference!. Proceedings: July 1, 2025 .No Proceedings: Aug 30, 2025. Link: .#ICCV2025
0
3
16
RT @ramoscsv: โ ๏ธ NEW DATES + NEW TRACK for LongVid-Foundations @ICCVConference!. Submit work & learn from leading experts: Katerina Fragkiaโฆ.
0
3
0
๐คฉ๐คฉ๐คฉ.
๐ข Announcing our 1st Workshop on Long Multi-Scene Video Foundations @ #ICCV2025 (@ICCVConference) in Honolulu, Hawaii!. Co-organized by Regev Cohen, @SivanDoveh, @hila_chefer , Jehanzeb Mirza, @hbXNov , @inbar_mosseri , Joao Magalhaes and me. website:
0
0
1
Nimrod will present liveXiv - an evolving dataset that tackles contamination - in Two days at @iclr_conf - super cool work and a super cool presenter ๐คฉ.
LiveXiv will be "live" on #ICLR2025 - Friday April 25th 10:00-12:30 Poster #356.@RGiryes @felipemaiapolo @LChoshen @WeiLinCV @jmie_mirza @leokarlin @ArbelleAssaf @SivanDoveh.
0
0
5
LiveXiv accepted to ICLR :) .It dynamically generates evolving benchmark from ArXiv to mitigate data contamination, ensuring ML models are evaluated on truly unseen data.
I am happy to share that LiveXiv accepted to ICLR 2025 ๐ฅณ.
0
1
6
RT @MMFMWorkshop: Excited to share that our 3rd Multimodal Workshop has been accepted to CVPR 2025 in Nashville! ๐ Looking forward to advanโฆ.
0
1
0
RT @EylonALevy: Itโs the middle of the night, day 440 of the October 7 War, and weโre still getting shot at by Iranโs pirate terrorists inโฆ.
0
77
0
Just back from NeurIPS where we presented 'ConMe', exploring how VLMs handle compositional reasoning. Loved catching up with old friends and making new connections. A perfect reminder that I should start planning for the May deadline! ๐
0
0
2
Results show better personalized localization without hurting prior performance. Annotate 1 image, and the model can annotate others!. Many thanks to all the exceptional collaborators!.Project page (paper & code):
1
1
5
We train models (e.g., LLaVA-OV, Qwen2VL) with instruction-tuning to improve context-aware localization. Using specific regularizations and personalised instructions generated from video tracking datasets, we shift VLMsโ focus from zero-shot to context-based learning.
1
0
3
Our work, "Teaching VLMs to Localize Specific Objects from In-context Examples," identifies this behaviour and takes steps towards correction. Given in-context samples of a specific object, the model will localize it in other (query) images.
1
0
3
Ever wanted to locate your cat in a database of images using just one reference image? Probably notโbut this highlights a gap in VLMs. They struggle to localize specific objects given in-context examples, often copying the last sample's location instead of learning from it.
1
7
39
ืชืงืืคื ืฉื ื ืืกืื ๐.
ืืขืจืืื ืืจืฅ ื ืืืจืช ืืืืฉ ืขื "ืืคืืจืืืืื". ืื ืื ืืืืืืช ๐. ืงืจืืื: Roni Harel
0
0
0