Deep Texture Encoding Network Training Example

In this tutorial, we show an example of training/testing Encoding-Net for texture recognition on MINC-2500 dataset. Comparing to original Torch implementation, we use different learning rate for pre-trained base network and the encoding layer (10x), disable color jittering after reducing lr and adopt much smaller training image size (224 instead of 352).

We provide the PyTorch Implementation of Encoding Layer at the GiHub repo. Make Sure to Install PyTorch Encoding before running the code.

Test Pre-trained Model

  • Clone the GitHub repo (you should have a copy during the installation)
      git clone
  • Download the MINC-2500 dataset to $HOME/data/minc folder. Download pre-trained model (training curve as shown below, model was pre-trained on train-1 split using single training size of 224, with an error rate of :math:19.98\% using single crop on test-1 set)
      cd PyTorch-Encoding/experiments
      bash model/

  • Test pre-trained model on MINC-2500::
      >>> python --dataset minc --model encodingnet --resume model/minc.pth.tar --eval
      # Teriminal Output:
      #[======================================== 23/23 ===================================>...]    Step: 104ms | Tot: 3s256ms | Loss: 0.719 | Err: 19.983% (1149/5750) 

Train Your Own Model

  • Example training command::
      python --dataset minc --model encodingnet --batch-size 64 --lr 0.01 --epochs 60 
  • Training options::
    -h, --help            show this help message and exit
    --dataset DATASET     training dataset (default: cifar10)
    --model MODEL         network model type (default: densenet)
    --widen N             widen factor of the network (default: 4)
    --batch-size N        batch size for training (default: 128)
    --test-batch-size N   batch size for testing (default: 1000)
    --epochs N            number of epochs to train (default: 300)
    --start_epoch N       the epoch number to start (default: 0)
    --lr LR               learning rate (default: 0.1)
    --momentum M          SGD momentum (default: 0.9)
    --weight-decay M      SGD weight decay (default: 1e-4)
    --no-cuda             disables CUDA training
    --plot                matplotlib
    --seed S              random seed (default: 1)
    --resume RESUME       put the path to resuming file if needed
    --checkname           set the checkpoint name
    --eval                evaluating

Extending the Software

This code includes an integrated pipeline and some visualization tools (progress bar, real-time training curve plots). It is easy to use and extend for your own model or dataset:

  • Write your own Dataloader to dataset/ folder

  • Write your own Model to model/ folder

  • Run the program::

    python –dataset mydataset –model mymodel


  • Hang Zhang, Jia Xue, and Kristin Dana. “Deep TEN: Texture Encoding Network.” The IEEE Conference on Computer Vision and Pattern Recognition (CVPR) 2017

      author = {Zhang, Hang and Xue, Jia and Dana, Kristin},
      title = {Deep TEN: Texture Encoding Network},
      booktitle = {The IEEE Conference on Computer Vision and Pattern Recognition (CVPR)},
      month = {July},
      year = {2017}
Written on October 17, 2017