Cloud Annotations Training

Custom Object Detection and Classification Training.

Quick & Dirty commands

It's recommended to go through one of the above walkthroughs, but if you already have and just need to remember one of the commands, here they are:

Project setup

git clone &&
cd training
  • classification
    python classification
  • object detection
    python object_detection

Training locally

python -m local.login
python -m
  • classification
    mkdir exported_graph
    python -m classification.retrain \
      --image_dir=.tmp/data \
      --saved_model_dir=exported_graph/saved_model \
      --tfhub_module= \
      --how_many_training_steps=500 \
  • object detection
    export PYTHONPATH=$PYTHONPATH:`pwd`/slim
    python -m object_detection.model_main \
      --pipeline_config_path=.tmp/pipeline.config \
      --model_dir=.tmp/checkpoint \
      --num_train_steps=500 &&
    python -m scripts.quick_export_graph

Training on IBM Cloud

python -m wml.login
python -m wml.start_training_run

Convert to desired format

python -m scripts.convert --tfjs --tflite --coreml