kaggle-global-wheat-detection

A collection of codes submitted for machine learning competitions on various platforms

kaggle-global-wheat-detection

Kaggle Competition - Global Wheat Detection - https://www.kaggle.com/c/global-wheat-detection

Preparing for Development

  1. Download All Dataset from competition
  2. Extract the data and create the structure as per follows in C:/kaggle directory

directory_structure

  1. Keep the pre_processed_train.csv in C:/kaggle/working/ Directory.

TO DO List:

Notebook Setup

  • Notebook Setup with Reading Dataset Functions

Pre Processing

  • Function = Pre Process.
  • Function = Split the bbox values to individual columns.
  • Function = Add images rows to df which are not having any bouding boxes but are there in train directory. Mark those bouding box values to 0.
  • Function = Add bbox_area column to df by calculating multiplication of bbox_width and bbox_height.
  • Function = Removes those rows from df which are having bbox_area more than given area_threshold value.

Data Augmentation

  • Function = Generate Flipped Horizontal Random Images and store to KAGGLE_WORKING directory and also append rows to df with the bbox values.
  • Function = Generate Flipped Vertical Random Images and store to KAGGLE_WORKING directory and also append rows to df with the bbox values.
  • Function = Generate some zoomed and cropped into 1024x1024 resolution images and store to KAGGLE_WORKING directory and also append rows to df with the bbox values.

Data Visualisation

  • Function = Create generic function to plot random image samples with bounding boxes

## References