Dynamically Editing Pipeline Config for Tensorflow Object Detection

Question

I'm using tensorflow object detection API, and I want to be able to edit config file dynamically in python, which looks like this. I thought of using protocol buffers library in python, but I'm not sure how to go about.

model {
ssd {
num_classes: 1
image_resizer {
  fixed_shape_resizer {
    height: 300
    width: 300
  }
}
feature_extractor {
  type: "ssd_inception_v2"
  depth_multiplier: 1.0
  min_depth: 16
  conv_hyperparams {
    regularizer {
      l2_regularizer {
        weight: 3.99999989895e-05
      }
    }
    initializer {
      truncated_normal_initializer {
        mean: 0.0
        stddev: 0.0299999993294
      }
    }
    activation: RELU_6
    batch_norm {
      decay: 0.999700009823
      center: true
      scale: true
      epsilon: 0.0010000000475
      train: true
    }
  }
 ...
 ...

}

Is there a simple/easy way to change specific values for fields like height in image_resizer -> fixed_shape_resizer from say 300 to 500? And write back the file with modified values without changing anything else?

EDIT: Though answer provided by @DmytroPrylipko worked for most of the parameters in the config, I face some issues with "composite field"..

That is, if we have configuration like:

train_input_reader: {
  label_map_path: "/tensorflow/data/label_map.pbtxt"
  tf_record_input_reader {
    input_path: "/tensorflow/models/data/train.record"
  }
}

And I add this line to edit input_path:

 pipeline_config.train_input_reader.tf_record_input_reader.input_path = "/tensorflow/models/data/train100.record"

It throws error:

TypeError: Can't set composite field

Answer 1

Yes, using Protobuf Python API is quite easy:

edit_pipeline.py:

import argparse

import tensorflow as tf
from google.protobuf import text_format
from object_detection.protos import pipeline_pb2


def parse_arguments():                                                                                                                                                                                                                                                
    parser = argparse.ArgumentParser(description='')                                                                                                                                                                                                                  
    parser.add_argument('pipeline')                                                                                                                                                                                                                                   
    parser.add_argument('output')                                                                                                                                                                                                                                     
    return parser.parse_args()                                                                                                                                                                                                                                        


def main():                                                                                                                                                                                                                                                           
    args = parse_arguments()                                                                                                                                                                                                                                          
    pipeline_config = pipeline_pb2.TrainEvalPipelineConfig()                                                                                                                                                                                                          

    with tf.gfile.GFile(args.pipeline, "r") as f:                                                                                                                                                                                                                     
        proto_str = f.read()                                                                                                                                                                                                                                          
        text_format.Merge(proto_str, pipeline_config)                                                                                                                                                                                                                 

    pipeline_config.model.ssd.image_resizer.fixed_shape_resizer.height = 300                                                                                                                                                                                          
    pipeline_config.model.ssd.image_resizer.fixed_shape_resizer.width = 300                                                                                                                                                                                           

    config_text = text_format.MessageToString(pipeline_config)                                                                                                                                                                                                        
    with tf.gfile.Open(args.output, "wb") as f:                                                                                                                                                                                                                       
        f.write(config_text)                                                                                                                                                                                                                                          


if __name__ == '__main__':                                                                                                                                                                                                                                            
    main() 

The way I call the script:

TOOL_DIR=tool/tf-models/research

(
   cd $TOOL_DIR
   protoc object_detection/protos/*.proto --python_out=.
)

export PYTHONPATH=$PYTHONPATH:$TOOL_DIR:$TOOL_DIR/slim

python3 edit_pipeline.py pipeline.config pipeline_new.config

Composite fields

In case of repeated fields, you must treat them as arrays (e.g. use extend(), append() methods):

pipeline_config.train_input_reader.tf_record_input_reader.input_path[0] = '/tensorflow/models/data/train100.record'

Eval Input reader error

This is a common error trying to edit the composite field. ( "no attribute tf_record_input_reader found" in case of eval_input_reader )

It's mentioned below in @latida's answer. Fix that by setting it as an array field.

pipeline_config.eval_input_reader[0].label_map_path  = label_map_full_path
pipeline_config.eval_input_reader[0].tf_record_input_reader.input_path[0] = val_record_path

Answer 2

This is the same above code with small changes that suit the tensorflow V2.

import argparse

import tensorflow as tf
from google.protobuf import text_format
from object_detection.protos import pipeline_pb2

def parse_arguments():                                                                                                                                                                                                                                                
    parser = argparse.ArgumentParser(description='')                                                                                                                                                                                                                  
    parser.add_argument('pipeline')                                                                                                                                                                                                                                   
    parser.add_argument('output')                                                                                                                                                                                                                                     
    return parser.parse_args()                                                                                                                                                                                                                                        


def main():                                                                                                                                                                                                                                                           
    args = parse_arguments()                                                                                                                                                                                                                                          
    pipeline_config = pipeline_pb2.TrainEvalPipelineConfig()                                                                                                                                                                                                          

    with tf.io.gfile.GFile(args.pipeline, "r") as f:                                                                                                                                                                                                                     
        proto_str = f.read()                                                                                                                                                                                                                                          
        text_format.Merge(proto_str, pipeline_config)                                                                                                                                                                                                                 

    pipeline_config.model.ssd.image_resizer.fixed_shape_resizer.height = 300                                                                                                                                                                                          
    pipeline_config.model.ssd.image_resizer.fixed_shape_resizer.width = 300                                                                                                                                                                                           

    config_text = text_format.MessageToString(pipeline_config) 
                                                                                                                                                                                                   
    with tf.io.gfile.GFile(args.output, "wb") as f:                                                                                                                                                                                                                
        f.write(config_text)                                                                                                                                                                                                                                          

if __name__ == '__main__':                                                                                                                                                                                                                                            
    main() 

Answer 3

I've found this to be a useful approach for overriding the object detection pipeline.config:

from object_detection.utils import config_util
from object_detection import model_lib_v2

PIPELINE_CONFIG_PATH = 'path_to_your_pipeline.config'

# Load the pipeline config as a dictionary
pipeline_config_dict = config_util.get_configs_from_pipeline_file(PIPELINE_CONFIG_PATH)

# OVERRIDE EXAMPLES
# Example 1: Override the train tfrecord path
pipeline_config_dict['train_input_config'].tf_record_input_reader.input_path[0] = 'your/override/path/to/train.record'
# Example 2: Override the eval tfrecord path
pipeline_config_dict['eval_input_config'].tf_record_input_reader.input_path[0] = 'your/override/path/to/test.record'

# Convert the pipeline dict back to a protobuf object
pipeline_config = config_util.create_pipeline_proto_from_configs(pipeline_config_dict)

# EXAMPLE USAGE:
# Example 1: Run the object detection train loop with your overrides (has to be string representation)
model_lib_v2.train_loop(config_override=str(pipeline_config))
# Example 2: Save the pipeline config to disk
config_util.save_pipeline_config(config, 'path/to/save/new/pipeline.config)