Upgrade to Pro
— share decks privately, control downloads, hide ads and more …
Speaker Deck
Features
Speaker Deck
PRO
Sign in
Sign up for free
Search
Search
Great Barrier Reef Model Pipeline: 15th place
Search
Maxwell
February 16, 2022
Science
1
120
Great Barrier Reef Model Pipeline: 15th place
https://www.kaggle.com/c/tensorflow-great-barrier-reef
All I want to use was YOLO-X!
Maxwell
February 16, 2022
Tweet
Share
More Decks by Maxwell
See All by Maxwell
Causal Impact -paper summary-
hoxomaxwell
2
470
Lecture materials at the University of Tokyo School of Medicine
hoxomaxwell
1
83
Kaggle Hungry Geese
hoxomaxwell
1
71
HuBMAP 17th place model pipeline
hoxomaxwell
1
63
LT: Shallow Dive into Bayes Factor
hoxomaxwell
6
1.2k
Kaggle APTOS 2019 @ U-Tokyo Med
hoxomaxwell
1
380
Cornell Birdcall 36th place solution
hoxomaxwell
2
200
Kaggle Bengali.AI 6 th place solution
hoxomaxwell
4
7.3k
Google Colaboratory Shortcuts
hoxomaxwell
2
970
Other Decks in Science
See All in Science
SIGDIAL論文読み会: PGTask: Introducing the Task of Profile Generation from Dialogues
kaiyo3
0
100
Machine Learning for Materials (Lecture 4)
aronwalsh
0
680
汎用原子シミュレータMatlantis のご紹介
matlantis
0
130
Onsager代数とその周辺 / Onsager algebra tsudoi
usamik26
0
390
Machine Learning for Materials (Lecture 7)
aronwalsh
0
730
Pandas 2 vs Polars vs Dask (PyDataGlobal 2023 December)
ianozsvald
0
450
MIKAMI Koichi
genomethica
0
180
(Forkwell Library #48)『詳解 インシデントレスポンス』で学び倒すブルーチーム技術
scientia
2
940
バックアップ『しながら』ランサムウェア検出も!? セキュリティ強化が満載 Veeam 12.1
climbteam
0
300
ABEMAの効果検証事例〜効果の異質性を考える〜
s1ok69oo
3
1.5k
AI(人工知能)の過去・現在・未来 —AIは人間を超えるのか—
tagtag
1
190
History towards Universal Neural Network Potential for Material Discovery
matlantis
0
150
Featured
See All Featured
For a Future-Friendly Web
brad_frost
172
9k
Optimizing for Happiness
mojombo
370
69k
Build The Right Thing And Hit Your Dates
maggiecrowley
25
2k
Creating an realtime collaboration tool: Agile Flush - .NET Oxford
marcduiker
14
1.5k
Making Projects Easy
brettharned
109
5.5k
Testing 201, or: Great Expectations
jmmastey
29
6.4k
A Philosophy of Restraint
colly
197
16k
Easily Structure & Communicate Ideas using Wireframe
afnizarnur
188
16k
Designing for humans not robots
tammielis
248
25k
Become a Pro
speakerdeck
PRO
12
4.5k
Ruby is Unlike a Banana
tanoku
96
10k
Happy Clients
brianwarren
92
6.4k
Transcript
Copyright 2022 Maxwell_110 Validation strategy - Sequence-based 4 fold CV
- The number of CoTS is close in each fold - Training data is frames with CoTs - Validation data includes frames w/o CoTs Resize up to 2.75 times using progressive learning 1280 720 Augmentation Increasing probability of applying augmentation as progressive learning progresses. - Default YOLO-X augmentations - random resize: (-5, 5) - mosaic / MixUp / hsv / flip: p = 0.6 -> 0.8 - degrees: Not used - translate: 0.1 - mosaic / MixUp scale: (0.5, 1.5) - RandomGamma - RGBShift - Sharpen - GaussNoise Batch Size: 4 GeForce RTX 3080 (x 2) Solution description in Kaggle discussion https://www.kaggle.com/c/tensorflow-great-barrier-reef/discussion/307691 Learning strategy - Progressive learning - Optimizer: default SGD (decay: 5e-4, momentum: 0.9) - LR: .000625 - Scheduler: yoloxwarmcos - min_lr_ratio: 0.1 - EMA: on - warmup_epochs: 5 - max_epoch: 30 TTA Seq-NMS https://arxiv.org/abs/1602.08465 https://github.com/tmoopenn/seq-nms n_frames: 2 confidence threshold: 0.07 linkage threshold: 0.1 nms th: 0.4 Weighted Box Fusion skip box threshold: 0.05 wbf IoU threshold: 0.45 Final confidence threshold: .08 Public LB : 0.607 Private LB : 0.714