Resnet_v2_50 on cat-vs-dog

The code uses predefined models from tensorflow's offical model zoo tensorflow slim.

To run it, you need to do the following steps

• download the dataset from kaggle, and extract it to data folder as below
• run create_label.py to generate label.csv, which contains the filename and it's label
• run train.py to train the model

python train.py --help
usage: A script to train resnet_2_50 [-h] [--batchsize BATCHSIZE] [--lr LR]
                                     [--numepochs NUMEPOCHS]
                                     [--testsize TESTSIZE]
                                     [--labelmap LABELMAP]
                                     [--numthreads NUMTHREADS]
                                     [--logdir LOGDIR]

optional arguments:
  -h, --help            show this help message and exit
  --batchsize BATCHSIZE
                        batch size
  --lr LR               learning rate
  --numepochs NUMEPOCHS
                        number of epochs to train
  --testsize TESTSIZE   ratio of validation data
  --labelmap LABELMAP   labelmap file
  --numthreads NUMTHREADS
                        number of threads to read data
  --logdir LOGDIR       log directory

structure of data directory

The directory of the dataset is discribed as below

• data
  • test1
    • 1.jpg
    • 2.jpg
    • ...
    • XXXX.jpg
  • train
    • cat.1.jpg
    • cat.2.jpg
    • ...
    • cat.XXXX.jpg
    • dog.1.jpg
    • dog.2.jpg
    • ...
    • dog.XXXX.jpg
  • sampleSubmission.csv