How can I start MobilNetSSD, How to Train?

[djmv]

How can I start MobilNetSSD, How to Train?

[sorelyss]

Hi, the short answer will be: there are a lot of tutorials of how to train SSD mobilnet using the Object Detection API of TensorFlow.

But if you want to try it quickly, you can follow these steps:

1. Clone the tensorflow/model repository @ anywhere you want.

2. Follow this instructions: https://github.com/tensorflow/models/blob/master/research/object_detection/g3doc/installation.md

2.1) Add to ~./bashrc:

export PYTHONPATH=${PYTHONPATH}:/PATH_TO/tensorflow/models/research:/PATH_TO/tensorflow/models/research/slim 

3. Download the checkpoint of a pretrained model (SSD, Mobilnet-SSD, FastRCNN...) here.

4. Replace PATH_TO_BE_CONFIGURED and number of classes at the model.config file.

---

5. Train!

#@/PATH_TO/tensorflow/models/research/
TRAIN_DIR=/PATH_TO_YOUR_TRAIN_FOLDER/ # for example: /home/username/Documents/project/train/
EVAL_DIR=/PATH_TO_YOUR_EVALUATION_FOLDER/
DETECT_DIR=PATH_TO_YOUR_DETECTION_FOLDER/output_inference_graph.pb
PIPELINE_CONFIG_PATH=/PATH_TO_YOUR_CONFIG_FILE/the_name.config

python3 object_detection/train.py --logtostderr \
    --pipeline_config_path=$PIPELINE_CONFIG_PATH \
    --train_dir=$TRAIN_DIR

5.1) Watch the training process

tensorboard --logdir=$TRAIN_DIR

---

References

• For creating your own .record files.
  https://github.com/tensorflow/models/blob/master/research/object_detection/g3doc/using_your_own_dataset.md
  http://www.machinelearninguru.com/deep_learning/tensorflow/basics/tfrecord/tfrecord.html
  https://planspace.org/20170323-tfrecords_for_humans/