Convert the scripts into a worker

Depends #5 (closed) #6 (closed) #7 (closed) #10 (closed)

Now that we have all the code needed, we can create a worker, similar to UFCNTrain. Create a new class PylaiaTrain(BaseWorker, TrainingMixin) in worker/train.py with the following structure:

• configure: mostly the same as what's been done with U-FCN up to the additional attributes definition (is_gpu_available, etc...)

• get_model_creation_config: method created in #5 (closed), currently hard-coded

• get_model_training_config: method created in #6 (closed), currently hard-coded

• retrieve_line_image: method created in #7 (closed), with the base-worker optimizations

• process_line: method created in #7 (closed)

• generate_training_dataset: method created in #7 (closed), will process on step ('train', 'val'), similar to build_dataset for U-FCN

• build_training_dataset: will call generate_training_dataset for each step, similar to build_training_dataset for U-FCN

• generate_syms: method created in #10 (closed)

• run

  • mktmpdir()
  • build_training_dataset()
    • Train folder
      • generate_training_dataset -> creates train.txt and corresponding images
    • Val folder
      • generate_training_dataset -> creates val.txt and corresponding images
  • generate_syms() -> syms.txt
  • get_model_creation_config
  • create_model # call to create_pylaia_model
  • get_model_training_config
  • train_model # call to train_pylaia_model
  • publish_model_version