Training

Download the dataset

We have prepared a script for downloading the training dataset. The script below downloads and merges the following datasets: MSRA-10K, DUT-OMRON, ECSSD, and FSS-1000.

# From CascadePSP/scripts/
python download_training_dataset.py

Note that the following script will create a dataset folder as follows:

+ CascadePSP/data/
  + DUTS-TE/
     - image_name_01.jpg
     - image_name_01.png
     - ...
  + DUTS-TR/
     - image_name_01.jpg
     - image_name_01.png
     - ...
  + ecssd/
     - image_name_01.jpg
     - image_name_01.png
     - ...
  + fss/
    + class_name/
         - image_name_01.jpg
         - image_name_01.png
     - ...
  + MSRA_10K/
     - image_name_01.jpg
     - image_name_01.png
     - ...

Running the training script

NOTE: Hyperparameters have been adjusted, and code are restructured. The new code yields slightly better performance with faster training time. Both the model used in the paper and the model trained with new code can be downloaded here.

Training can be done with following command with some distinguishable id:

# From CascadePSP/
python train.py some_unique_id

Note that you can change the hyperparameter by specifying arguments, e.g. to change batch size.

# From CascadePSP/
python train.py -b 10 some_unique_id

Please check hyper_para.py for more options.

After training

Tensorboard log file will be stored in CascadePSP/log/some_unique_id_timestamp

Model will be saved in CascadePSP/weights/some_unique_id_timestamp