Kaggle Humpback Whale Identification Challenge 1st place code

Recent Updates

[2019.3.1 16:44] uploading mask file

Dependencies

pip install opencv-python

• python==3.6
• torch==0.4.1
• torchvision==0.2.1
• pandas
• tqdm
• opencv-python

Solution

https://www.kaggle.com/c/humpback-whale-identification/discussion/82366

prepare data

competition dataset

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

I made links.

ln -s /data/kaggle/comp/train train
ln -s /data/kaggle/comp/test test

mask

cd input
unzip mask.zip unzip model_50A_slim_ensemble.csv.zip (Originally downloaded from https://drive.google.com/file/d/1hfOu3_JR0vWJkNlRhKwhqJDaF3ID2vRs/view?usp=sharing)

playground data

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

Train

line 301 in train.py
step 1.
       freeze = False
       model_name = 'senet154'
       min_num_class = 10
       checkPoint_start = 0
       lr = 3e-4
       #until train map5 >= 0.98

step 2.
       freeze = True
       model_name = 'senet154'
       min_num_class = 0
       checkPoint_start = best checkPoint of step 1
       lr = 3e-4

step 3.
       freeze = True
       model_name = 'senet154'
       min_num_class = 0
       checkPoint_start = best checkPoint of step 2
       lr = 3e-5

Test

line 99 in test.py
       checkPoint_start = best checkPoint of step 3

Questions

• Did you manually create the bboxes? (See inputs/bboxs.csv)