M.Bülent Sarıyıldız
@mbsariyildiz
Followers
463
Following
2K
Statuses
175
Research scientist at @naverlabseurope
Grenoble, France
Joined January 2013
Want to distill diverse visual encoders into a universal one that outperforms all original ones? Then check out our #ECCV2024 paper! UNIC: Universal Classification Models via Multi-teacher Distillation A quick summary 🧵 (1/6)
2
23
78
RT @dlarlus: 📢 Job Alert 📢 @naverlabseurope My team is looking for a Research Scientist in Visual Representation Learning - More info http…
0
38
0
0
4
24
@mdorkenw @eccvconf @imisra_ @oriane_simeoni @endernewton @olivierhenaff @y_m_asano @YutongBAI1002 Final talk of the workshop by @YutongBAI1002 on bottom up visual learning as for SSL 🎊
0
1
7
@mdorkenw @eccvconf @imisra_ @oriane_simeoni @endernewton @olivierhenaff @y_m_asano @YutongBAI1002 We are on level-0. Escalator is not working unfortunately, take the lift on the left of the entrance.
0
0
2
RT @GalloDanilo_: We are very happy to announce the 2nd edition of our HRI Symposium! 🎉 Looking forward to welcoming everyone in Grenoble (…
0
2
0
@PavloMolchanov By the way, we provide a comparison to RADIO (when trained on IN-1K) in Table-2. Also available in this thread -
We extend our study to larger teachers trained on arbitrary datasets: MetaCLIP ViT-H and DINOv2 ViT-G, and show that UNIC practically matches the performance of these two state-of-the-art foundation models with a smaller ViT-Large encoder. (4/6)
0
0
1
Paper: Code and models: Work by @skamalas, @dlarlus, @WeinzaepfelP, @ThomasLUC4S and myself at @naverlabseurope (6/6)
0
1
13