Quantization with openvino_2021.1 (#9)

README.md

@@ -61,7 +61,7 @@ python3 evaluate.py --engine opvn --xml resnet_v2_101_299_opt.xml --dataset ILSV
 To quantizes the model to int8 model run:
 
 ```bash
-python3 calibration.py --xml resnet_v2_101_299_opt.xml --data ILSVRC2012_img_val --annotation ILSVRC2012_img_val/val.txt
+python3 quantize.py --xml resnet_v2_101_299_opt.xml --data ILSVRC2012_img_val --annotation ILSVRC2012_img_val/val.txt
 ```
 
 This script created ```/model/optimised``` folder in current folder with quantized model.

calibration.py → quantize.py

@@ -5,9 +5,9 @@
 
 from addict import Dict
 from compression.graph import load_model, save_model
-from compression.data_loaders.data_loader import DataLoader
+from compression.api.data_loader import DataLoader
 from compression.engines.ie_engine import IEEngine
-from compression.custom.metric import Metric
+from compression.api.metric import Metric
 from compression.pipeline.initializer import create_pipeline
 
 parser = argparse.ArgumentParser(description="Quantizes OpenVino model to int8.",
@@ -111,39 +111,55 @@ def higher_better(self):
  def get_attributes(self):
  return {}
 
+# Dictionary with the FP32 model info
 model_config = Dict({
  'model_name': argv.model_name,
  "model": argv.xml,
  "weights": argv.xml.split('.xml')[0] + '.bin'
 })
-engine_config = {
+
+# Dictionary with the engine parameters
+engine_config = Dict({
  'device': 'CPU',
  'stat_requests_number': 4,
  'eval_requests_number': 4
-}
-dataset_config = {
+})
+
+# Dictionary witn input dataset info
+dataset_config = Dict({
  'data_source': argv.data,
  'annotation_file': argv.annotation,
-}
+})
+
+# Quantization algorithm settings
 algorithms = [
  {
- 'name': 'DefaultQuantization',
+ 'name': 'DefaultQuantization', # Optimization algorithm name
  'params': {
  'target_device': 'CPU',
- 'preset': 'performance',
- 'stat_subset_size': 300
+ 'preset': 'performance', # Preset [performance (default), accuracy] which controls the quantization mode 
+ # (symmetric and asymmetric respectively)
+ 'stat_subset_size': 300 # Size of subset to calculate activations statistics that can be used
+ # for quantization parameters calculation.
  }
  }
 ]
 
+# Load the model.
 model = load_model(model_config)
 
+# Initialize the data loader and metric.
 data_loader = DatasetsDataLoader(dataset_config)
 metric = AccuracyMetric()
 
-loss = None
-engine = IEEngine(engine_config, data_loader, metric, loss)
+# Initialize the engine for metric calculation and statistics collection.
+engine = IEEngine(engine_config, data_loader, metric)
+
+# Initialize the engine for metric calculation and statistics collection.
 pipeline = create_pipeline(algorithms, engine)
 
+# Execute the pipeline.
 compressed_model = pipeline.run(model)
+
+# Save the compressed model.
 save_model(compressed_model, argv.int8_dir)