CNN_project

EC601 project2

A simple image recognition classfier using CNN, Keras and Tensorflow backend on Google Colab

Dataset

Flower Recognition (Kaggle) (only use daisy and dandelion)
Test data are downloaded randomly from Google image

Directory

/content/drive/My Drive/flowers/
    train/
        daisy/
            daisy1.jpg
            daisy2.jpg
            ...
        dandelion/
            dandelion1.jpg
            dandelion2.jpg
            ...
    validation/
        daisy/
            daisy1.jpg
            daisy2.jpg
            ...
        dandelion/
            dandelion1.jpg
            dandelion2.jpg
            ...  
/content/drive/My Drive/flowers_test/
    train/
        daisy/
            daisy1.jpg
            daisy2.jpg
            ...
        dandelion/
            dandelion1.jpg
            dandelion2.jpg
            ...
    validation/
        daisy/
            daisy1.jpg
            daisy2.jpg
            ...
        dandelion/
            dandelion1.jpg
            dandelion2.jpg
            ...

Run

1. Download the dataset and save in Googledrive
2. Create a Colab notebook in your Googledrive
3. Enable the TPU accelerator
4. Run

three layer CNN Result

val_acc: 0.8646

Filename Predictions
daisy/daisy1.jpg 0
daisy/daisy2.jpg 0
daisy/daisy3.jpg 0
daisy/daisy4.jpg 0
daisy/daisy5.jpg 0
daisy/daisy6.jpg 1
daisy/daisy7.jpg 0
daisy/daisy8.jpg 0
dandelion/dandelion.jpg 1
dandelion/dandelion1.jpg 1
dandelion/dandelion2.jpg 1
dandelion/dandelion3.jpg 1
dandelion/dandelion4.png 1
dandelion/dandelion5.jpg 1
dandelion/dandelion6.jpg 1
dandelion/dandelion7.jpg 1

Resnet Result

val_acc: 0.7717

Filename Prediction
daisy/daisy1.jpg 1
daisy/daisy2.jpg 1
daisy/daisy3.jpg 0
daisy/daisy4.jpg 0
daisy/daisy5.jpg 0
daisy/daisy6.jpg 0
daisy/daisy7.jpg 0
daisy/daisy8.jpg 0
dandelion/dandelion.jpg 1
dandelion/dandelion1.jpg 1
dandelion/dandelion2.jpg 1
dandelion/dandelion3.jpg 1
dandelion/dandelion4.png 1
dandelion/dandelion5.jpg 1
dandelion/dandelion6.jpg 1
dandelion/dandelion7.jpg 1