soumya997 / siim-isic-melanoma-classification-kaggle

Kaggle challenge for Melanoma Detection.

SIIM-ISIC-Melanoma-classification-Kaggle-challange

plotted images

1. is malignent images,
2. is benign images,
3. inpainted images(after hair removal)
4. anatom_site_general_challenge pi chart,
5. count of age approx,
6. count of diagnosis and many more in the notebooks.

Model structure for VGG:

model.add(VGG19(include_top=False, weights='imagenet', input_shape= inputShape))
model.add(Flatten())
model.add(Dense(32))
model.add(LeakyReLU(0.001))
model.add(Dense(16))
model.add(LeakyReLU(0.001))
model.add(Dense(1, activation='sigmoid'))

ResNet101 Model summary:

Model: "sequential"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
resnet101 (Model)            (None, 7, 7, 2048)        42658176  
_________________________________________________________________
global_average_pooling2d (Gl (None, 2048)              0         
_________________________________________________________________
dense (Dense)                (None, 1)                 2049      
=================================================================
Total params: 42,660,225
Trainable params: 42,554,881
Non-trainable params: 105,344
_________________________________________________________________

File structure:

1. ResNet101 training inside resnet101-with-focal-loss-and-img-aug.ipynb
2. VGG16 training is in baseline-submission-keras-vgg16
3. Image Analysis is in cancer-detection-analysis.ipynb
4. Tabular data analysis is in eda-w-plotly-and-stacking-on-tabular-data-0-685.ipynb

Future work:

1. woking on the image analysis more
2. woking on bettering the model performance.