Train SSD

This page shows how to train SSD with your own dataset.

SSD is a neural network model used for object detection.

Available Dataset Class

DetectionDataset (imported from jsk_recognition_utils.datasets)

This class assumes the following directory structure for each split.

path_to_awesome_dataset/
|-- JPEGImages
|   |-- foo.jpg
|   |-- bar.jpg
|   `-- etc.
|-- SegmentationClass
|   |-- foo.npy
|   |-- bar.npy
|   `-- etc.
|-- SegmentationObject
|   |-- foo.npy
|   |-- bar.npy
|   `-- etc.
|-- class_names.txt
`-- etc.

BboxDetectionDataset (imported from jsk_recognition_utils.datasets)

This class assumes the following directory structure for each split. This can be generated using labelme

path_to_awesome_dataset/
|-- JPEGImages
|   |-- foo.jpg
|   |-- bar.jpg
|   `-- etc.
|-- Annotations
|   |-- foo.xml
|   |-- bar.xml
|   `-- etc.
|-- AnnotationsVisualization (not necessary)
|   |-- foo.jpg
|   |-- bar.jpg
|   `-- etc.
`-- class_names.txt

Arguments

• --train-dataset-dir (string, default: $(rospack find jsk_perception)/learning_datasets/kitchen_dataset/train)

• --val-dataset-dir (string, default: $(rospack find jsk_perception)/learning_datasets/kitchen_dataset/test)

  Directory name which contains dataset for training and validation respectively.

• --dataset-type (string, default: instance)

  Choose dataset type. If you use DetectionDataset structure, choose instance. If you use BboxDetectionDataset, choose bbox.

• --model-name (string, default: ssd512)

  Model name. Currently, ssd300 and ssd512 are supported.

• --gpu (int, default: 0)

  GPU id. -1 means CPU mode, but we recommend to use GPU for much faster computing.

• --batch-size (int, default: 8)

  Number of images used simultaneously in each iteration.

  You should decrease this number when you face memory allocation error.

• --max-epoch (int, default: 100)

  Stop trigger for training.

• --out-dir (string, default: ${ROS_HOME}/learning_logs/<timestamp>)

  Output directory name.

Output

All these files will be automatically generated under <out_dir>.

• log.json

• model_snapshot.npz

Usage

rosrun jsk_perception train_ssd.py [ARGS]

Sample usage with pre-trained model

There are some pre-trained SSD model on jsk_perception.Getting trained data by build jsk_perception or run script install_trained_data

You can use 73b2 kitchen model with jsk_perception node.

roslaunch jsk_perception sample_ssd_object_detector_73b2_kitchen.launch

Sample Output

73b2 kitchen model is some of the typical example of pre-trained SSD model on jsk_perception.The results of SSD using 73b2 kitchen model are as follows.