/bird_detection_model

Primary LanguageJupyter Notebook

Initialization

Environment setup

install all the environment based on the environment.yml file using conda or others

for conda user can run these code.

conda env create -f environment.yml -p ./env

Dataset Download

  1. obtain your kaggle api from kaggle website.
  2. open eda.ipynb run all commands and enter your kaggle's username and password.

Preprocessing and Model Building

  1. Clone the sample.ipynb with your name <name>.ipynb.
  2. Code review, the first section is data preprocessing, involves
    • Resizing
    • Normalization [0, 255] -> [0, 1]
    • Data Augmentation i. horizontal flip ii. rotation
  3. test all functions is working, run all the entire file
  4. if no error, remember search 'REMARK' and comment out the testing part which is reducing the dataloader for testing purpose
### REMARK: for testing purposes, reduce the dataset to 10% of the original size
train_dataset.dataset.samples = train_dataset.dataset.samples[:int(len(train_dataset)*0.1)]
test_dataset.dataset.samples = test_dataset.dataset.samples[:int(len(test_dataset)*0.1)]
valid_dataset.dataset.samples = valid_dataset.dataset.samples[:int(len(valid_dataset)*0.1)]

len(train_dataset), len(test_dataset), len(valid_dataset)
  1. remember save every trained model and history, screenshot all the training and evaluation as well.
  2. remember rename each trained model version to avoid overwritting.
  3. run tensorboard --logdir=runs in terminal to view the accuracy and loss result.