/whale_pytorch

Primary LanguageJupyter Notebook

Kaggle Sperm Whale Identification Challenge

docithurtswhaleIP code

Dependencies

  • python==3.6
  • torch==0.4.1
  • torchvision==0.2.1

other dependencies in requierements.txt

Solution

Forked and Adapted from: https://github.com/earhian/Humpback-Whale-Identification-1st- https://www.kaggle.com/c/humpback-whale-identification/discussion/82366

prepare data

competition dataset

trainset -> ./WC_input/train
testset -> ./WC_input/test

mask

cd input
Download training masks https://storage.googleapis.com/kaggle-forum-message-attachments/459392/11072/masks.zip unzip mask.zip

playground data

download humpback whale kaggle data, then put them into input
https://www.kaggle.com/c/whale-categorization-playground/data

download sperm whale data, flatten the data in single directories train, test, and data (all) then put them into WC_input https://gdsc.ce.capgemini.com/resources/

Train Hyperparams

line 345 in train.py
step 1.
       freeze = False
       model_name = 'seresnext101'
       min_num_class = 10
       checkPoint_start = 0
       lr = 3e-4
       #until train map5 >= 0.98

step 2.
       freeze = True
       model_name = 'seresnext101'
       min_num_class = 0
       checkPoint_start = 0
       lr = 3e-4

step 2.
       freeze = True
       model_name = 'seresnext101'
       min_num_class = 0
       checkPoint_start = 0
       lr = 3e-4
       test_train=True

Test

line 130 in test.py
       checkPoint_start = best checkPoint of step 2