aakanksha-ns / wheat-detection

Kaggle competition - https://www.kaggle.com/c/global-wheat-detection

GLOBAL WHEAT DETECTION

CONTRIBUTORS

• Akanksha .
• Aakanksha Nallabothula Surya

TECHNOLOGIES USED:

• PyTorch
• Deep Learning - Faster-RCNN

GOAL

To predict bounding boxes around each wheat head in images that have them.

Kaggle competition link: https://www.kaggle.com/c/global-wheat-detection

DATA

Training data consists of 3422 image files, totaling around 613 Mb. Train.csv has columns such as image_id, width and height of the image, and bounding box dimension [xmin, ymin, width, height]

FINAL REPORT

https://github.com/aakanksha-ns/wheat-detection/blob/master/Deep_Learning_Project_Report.pdf

CODE

Exploratory Data Analysis:
https://github.com/aakanksha-ns/wheat-detection/blob/master/Wheat_Detection_EDA.ipynb

Faster RCNN attempt 1:
https://github.com/aakanksha-ns/wheat-detection/blob/master/FasterRCNN_attempt1.ipynb

Final Training notebook:
https://github.com/aakanksha-ns/wheat-detection/blob/master/faster_r-CNN_resnet152_train.ipynb

Final Inference notebook:
https://github.com/aakanksha-ns/wheat-detection/blob/master/faster_r-CNN_Inference.ipynb