#DSB2018 - Twitter Hashtag

almost 8 years ago

Auch die Drogenbeauftrage der Bundesregierung #MarleneMortler bestätigt im aktuellen Drogen- und Suchtbericht #DSB2018, dass E-Zigaretten deutlich weniger Schadstoffe als herkömmliche Zigaretten aufweisen: https://t.co/guQBaxT1QL

0

4

3

0

Kim Sineke @KimSineke

almost 8 years ago · Sandton

Fully In! #DSB2018 #DubaneSpringBreak18 🏖💦🏄🏾‍♀️🤸🏾‍♀️🌊☀️

0

2

1

0

bia @beatijolo

almost 8 years ago · Armação dos Búzios

DSB 2018 ❤ Dias incríveis, com uma equipe sensacional, com o baby Verneque. Já quero de novo ☀️🏆 #dsb2018 #verneque #barco #catamarã #buzios #rj #equipe #voltadsb em Armação Dos Búzios,… https://t.co/iMtg37OHSb

0

1

0

Pierre Deep Sea @pgirguis

almost 8 years ago

I love this conference. A colleague, whom I ran into while she was stepping out of the poster session for a minute said with a smile “I had to get out of there for a bit. It’s loud, warm, and full of social anxiety.” Man that had me rolling. #DSB2018 #geekhumor #rollingwithit

1

8

1

0

Fabio Cortesi @FCortesi

almost 8 years ago

Yayyy more fishy talks #DSB2018 @TiffoFisho assessing the importance of depth for fish communities 🎣 #fishsci #deepsea @DSBSoc

FCortesi's tweet photo. Yayyy more fishy talks #DSB2018 @TiffoFisho assessing the importance of depth for fish communities 🎣 #fishsci #deepsea @DSBSoc https://t.co/n83NNRcPEX

0

4

1

0

Alex Kalinin @alxndrkalinin

about 8 years ago

Evaluation of Deep Learning Strategies for Nucleus Segmentation in Fluorescence Images https://t.co/O9smSEIGSr - 23,165 manually labeled nuclei - predicting masks & boundaries - @CellProfiler vs DeepCell vs U-Net (best perf) - amount & variety of data are important - pre-#DSB2018

alxndrkalinin's tweet photo. Evaluation of Deep Learning Strategies for Nucleus Segmentation in Fluorescence Images https://t.co/O9smSEIGSr
- 23,165 manually labeled nuclei
- predicting masks & boundaries
- @CellProfiler vs DeepCell vs U-Net (best perf)
- amount & variety of data are important
- pre-#DSB2018 https://t.co/CvtKNCDnFN

0

59

30

4

0

BIRDS Project @birdsproject

about 8 years ago

#DSB2018 was held at Helsinki last week. Participants share a great research and networking atmosphere. You can find here more info about the workshop: https://t.co/15R5hlLEKp (photo by Tuukka Norri)

birdsproject's tweet photo. #DSB2018 was held at Helsinki last week. Participants share a great research and networking atmosphere. You can find here more info about the workshop: https://t.co/15R5hlLEKp (photo by Tuukka Norri) https://t.co/u9d5f8bMBx

0

2

0

Одна інтеллектуальна одиниця @cvtalks

about 8 years ago

10th place, gold medal, top 1% on leaderboard in #kaggle #dsb2018. https://t.co/cumr10TOQq. And my deep learning background is only 5 months so far ^_^. Thanks to my teammates and #keras & #pytorch for such an amazing frameworks.

1

4

0

Angel Lopez-Urrutia @LopezUrrutiaA

about 8 years ago

I'm thrilled to have finished third in the 2018 @kaggle data science bowl. #dsb2018 #datascibowl https://t.co/KpMerAvq4o

0

1

0

Alex Kalinin @alxndrkalinin

over 8 years ago

Description of 4th place solution for #Kaggle Data Science Bowl 2018 https://t.co/lViYL1K87y - inspired by Deep Watershed Transform - 2 stacked U-Nets trained from scratch - 1st UNet: vector fields - 2nd: mask, watershed levels & nuclei centers - heuristic postprocessing #dsb2018

alxndrkalinin's tweet photo. Description of 4th place solution for #Kaggle Data Science Bowl 2018 https://t.co/lViYL1K87y
- inspired by Deep Watershed Transform
- 2 stacked U-Nets trained from scratch
- 1st UNet: vector fields
- 2nd: mask, watershed levels & nuclei centers
- heuristic postprocessing
#dsb2018 https://t.co/YCMn5pLKvX

0

91

32

2

0

Alexandre Cadrin-Chênevert

@alexandrecadrin

over 8 years ago

@judywawira @jeremyphoward @fastdotai @YamashitaRikiya For #DSB2018, maybe about 10/week between 1-24 hours each. Initializing the base CNN weights with imagenet/coco of Mask-RCNN was speeding up experiments. Model training doesn't use much human time if not babysitted. Image post-processing was more time consuming to code.

0

3

1

0

Alexandre Cadrin-Chênevert

@alexandrecadrin

over 8 years ago

@judywawira #DSB2018 was a very competitive data science challenge with 3634 teams. As a full-time attending radiologist with an active family, I am very happy with our top 6% result with @jamesrequa using a Mask-RCNN network. Best of all, I learned a lot as usual from others participants!

3

33

7

0

Alex Kalinin @alxndrkalinin

over 8 years ago

Applying Deep Watershed Transform to @Kaggle Data Science Bowl 2018 https://t.co/6H9arsCkBK Dockerized #PyTorch code: https://t.co/NCs7Hz4x0Y - U-Net with VGG16 encoder (aka TernausNet) - deep watershed-like solution with learned energy - lots of augmentations #dsb2018

alxndrkalinin's tweet photo. Applying Deep Watershed Transform to @Kaggle Data Science Bowl 2018 https://t.co/6H9arsCkBK
Dockerized #PyTorch code: https://t.co/NCs7Hz4x0Y
- U-Net with VGG16 encoder (aka TernausNet)
- deep watershed-like solution with learned energy
- lots of augmentations
#dsb2018 https://t.co/WZd3yubxqU

3

152

57

8

0

Alex Kalinin @alxndrkalinin

over 8 years ago

Summary of winning solution for @kaggle Data Science Bowl 2018 https://t.co/g5FAPj0Wwd - modified U-Net > Mask-RCNN - add borders btw cells as targets - heavy augmentations - deep encoders: DPN-92, Resnet-152, InceptionResnetV2 - watershed + morphology for postprocessing #dsb2018