Implementation of camera model identification models in kaggle .

The repository contains the implmentation of various image classification models like ResNet,DPN,DenseNet,etc in PyTorch deep learning framework .

models in this repo:

• densenet121
• densenet161
• densenet169
• densenet201
• resnet50
• resnet101
• resnet152
• resnext101
• dpn92
• dpn98
• se_resnet50
• se_resnet101
• se_resnext50

To train models and get predictions the following is required:

• Python 3.6

packages:

• torch==1.0.0
• torchvision==0.2.1
• scipy==1.0.0
• tqdm==4.28.1
• tensorboardX==1.4
• matplotlib==2.1.2
• opencv-python==3.4.2.17
• numpy==1.14.0
• pandas==0.20.3

1. Install packages with pip install -r requirements.txt
2. Download dataset from kaggle
3. Place train dataset from Kaggle competition to dataset/train. Place test dataset from Kaggle competition to dataset/test.
4. run python data_split.py
5. run python runTrain.py --batch_size=64 --pretrained=True --learning_rate=0.0001 --epochs=100
6. run python runMakePseudo.py --batch_size=128
7. run python runTrain.py --batch_size=64 --learning_rate=0.0001 --epochs=40 --resume=True --use_pseudo=True
8. run python runTest.py --batch_size=128