Coral Classification Challenge
'%3e%3cpath fill='%23001489' d='M0 0v6h9V0z'/%3e%3cpath fill='%23e03c31' d='M0 0v3h9V0z'/%3e%3cg stroke='%23fff' stroke-width='2'%3e%3cpath id='d' d='m0 0 4.5 3L0 6m4.5-3H9'/%3e%3cuse xlink:href='%23b' stroke='%23ffb81c' clip-path='url(%23c)'/%3e%3c/g%3e%3cuse xlink:href='%23d' fill='none' stroke='%23007749' stroke-width='1.2'/%3e%3c/g%3e%3c/svg%3e)
Helping South Africa
$800 USD
Challenge completed ~4 years ago
Classification
Computer Vision
139 joined
44 active
Start
Oct 27, 21
Close
Oct 31, 21
Reveal
Oct 31, 21
1st Place Approach
Notebooks · 2 Nov 2021, 12:24 · edited 4 months later · 1
All thanks to @ZINDI for hosting this challenge,I was able to secure 1st place in this interesting competition.
I will not share my solution code, But instead of I will share my winning approach and some repositories that help you to code my solution.
--------------------------------------------------------------------------
Apporach
My winning solution is a weighted average between ResNet18 - ResNet50 - EfficientNet-B3.
I used some augmentations to deal with the bad image quality such as Histogram Equalization, and some classic augmentations such as Rotation, Flip ...
My Overall CV was 0.6607 [ using 5 K-Folds ] , giving 0.665 in public LB and 0.646 in public LB.
--------------------------------------------------------------------------
Github Repositories to learn
- this repo contains Both Pytorch code [check densenet169 - densenet201 - resnext101 resnext50 Folders ] and 2 Optimized Fastai Codes .
- For those who want to learn coding in PyTorch, please follow the amazing Abhishek Thakur Youtube Channel
Interesting, my solution was quite simple. No more than 5 lines of code. i used Resnet101 implemented with Fastai. The images were terrible. My model tended to perform better without augmentation. i got 0.663.. on public leaderboard and 0.638 on private leaderboard.