Skip to content

Image Segmentation Keras

Image Segmentation Keras is a Keras/Tensorflow based image segmentation framework. Custom docker images with additional tools are available from here:

https://github.com/waikato-datamining/tensorflow/tree/master/image-segmentation-keras

Prerequisites#

Make sure you have the directory structure created as outlined in the Prerequisites.

Data#

In this example, we will use the CamVid-12 dataset, which consists of still images from dashcam videos from a city environment (12 different labels).

Download the dataset from the following URL into the data directory and extract it:

https://datasets.cms.waikato.ac.nz/ufdl/data/camvid12/camvid12-grayscale.zip

Once extracted, rename the grayscale directory to camvid-grayscale.

Now we have to convert the format from grayscale into blue channel, which the framework uses. We can do this by using the image-dataset-converter library. At the same time, we can split the dataset into train, validation and test subsets.

From within the applied_deep_learning directory, run the following command:

docker run --rm -u $(id -u):$(id -g) \
  -v `pwd`:/workspace \
  -t waikatodatamining/image-dataset-converter:0.0.4 \
  idc-convert \
    -l INFO \
    from-grayscale-is \
      -i "/workspace/data/camvid-grayscale/*.png" \
      --labels sky building pole road pavement tree signsymbol fence car pedestrian bicyclist unlabelled \
    to-blue-channel-is \
      -o /workspace/data/camvid-bluechannel-split \
      --split_names train val test \
      --split_ratios 70 15 15

Training#

For training, we will use the following docker image:

waikatodatamining/image-segmentation-keras:1.14.0_0.3.0

The training script is called keras_seg_train, for which we can invoke the help screen as follows:

docker run --rm -t waikatodatamining/image-segmentation-keras:1.14.0_0.3.0 keras_seg_train --help

It is good practice creating a separate sub-directory for each training run, with a directory name that hints at what dataset and model were used. So for our first training run, which will use mainly default parameters, we will create the following directory in the output folder:

camvid12-keras-unet50

Image Segmentation Keras does not use a config file, but can be configured via command-line parameters instead:

  • the number of classes (--n_classes) is actual classes plus background
  • the width/height of the inputs must be multiples of 32 (--input_height, --input_width)
  • the output directory is specified via --checkpoints_path (make sure to have a trailing slash!)
  • in our case, the images and the annotations reside in the same directories, so --train_images and --train_annotations point to the same directory (analog for --val_images and --val_annotations)
  • with --epochs you can specify for how long the model will get trained
  • --model_name specifies the architecture and backend (see here for available options)

Kick off the training of a U-Net with a ResNet50 backend using the following command:

docker run --rm \
  -u $(id -u):$(id -g) \
  --gpus=all \
  -v `pwd`:/workspace \
  -v `pwd`/cache:/tmp/.keras \
  -t waikatodatamining/image-segmentation-keras:1.14.0_0.3.0 \
  keras_seg_train \
  --checkpoints_path /workspace/output/camvid12-keras-unet50/ \
  --train_images /workspace/data/camvid-bluechannel-split/train \
  --train_annotations /workspace/data/camvid-bluechannel-split/train \
  --val_images /workspace/data/camvid-bluechannel-split/val \
  --val_annotations /workspace/data/camvid-bluechannel-split/val \
  --epochs 10 \
  --n_classes 13 \
  --input_height 384 \
  --input_width 480 \
  --model_name resnet50_unet

Predicting#

Using the keras_seg_poll script, we can batch-process images placed in the predictions/in directory as follows (e.g., from our test subset): 

docker run --rm \
  -u $(id -u):$(id -g) \
  --gpus=all \
  -v `pwd`:/workspace \
  -v `pwd`/cache:/tmp/.keras \
  -t waikatodatamining/image-segmentation-keras:1.14.0_0.3.0 \
  keras_seg_poll \
  --checkpoints_path /workspace/output/camvid12-keras-unet50/ \
  --prediction_in /workspace/predictions/in \
  --prediction_out /workspace/predictions/out

Notes

  • The predictions are output in Indexed PNG format.

  • You can view the predictions with the ADAMS Preview browser:

Example prediction

Screenshot

Screenshot