This repository demonstrates web-based implementation of DeOldify, a Deep Learning based project for colorizing and restoring old images. This project demonstrates implementation of 2 models:
- Original artistic DeOldify model ("original" folder)
- Quantized DeOldify model ("quantized" folder).
To use a DeOldify example, copy the corresponding html file contents. To dowload and locally serve models download them from the link provided:
- Original artistic model: https://cdn.glitch.me/2046b88b-673a-457f-b1b8-7169ce9bf13a/deoldify-art.onnx (~243mb)
- Quantized model: https://cdn.glitch.me/2046b88b-673a-457f-b1b8-7169ce9bf13a/deoldify-quant.onnx (~61mb)
Original onnx files were taken from releases page of Deoldify Onnx repository by Thomas De. To quantize an onnx file do the following:
- Open Google Colab and create a new Notebook.
- Upload onnx file downloaded from releases page above. Also upload "remove_initializer_from_input.py". The file content is given below:
# /content/remove_initializer_from_input.py
import argparse
import onnx
def get_args():
parser = argparse.ArgumentParser()
parser.add_argument("--input", required=True, help="input model")
parser.add_argument("--output", required=True, help="output model")
args = parser.parse_args()
return args
def remove_initializer_from_input():
args = get_args()
model = onnx.load(args.input)
if model.ir_version < 4:
print("Model with ir_version below 4 requires to include initilizer in graph input")
return
inputs = model.graph.input
name_to_input = {}
for input in inputs:
name_to_input[input.name] = input
for initializer in model.graph.initializer:
if initializer.name in name_to_input:
inputs.remove(name_to_input[initializer.name])
onnx.save(model, args.output)
if __name__ == "__main__":
remove_initializer_from_input()- Run the following code:
# Install dependencies and
!pip install onnxruntime
!pip install onnxRun onnx preprocess:
!python -m onnxruntime.quantization.preprocess --input '/content/deoldify.onnx' --output '/content/deoldify-final.onnx'Generate quantized file:
import onnx
from onnxruntime.quantization import quantize_dynamic, QuantType
model_fp32 = '/content/deoldify-final.onnx'
model_quant = '/content/deoldify-quant.onnx'
quantized_model = quantize_dynamic(model_fp32, model_quant, weight_type=QuantType.QUInt8)Remove initializer from input of deoldify-quant.onnx file (otherwise the model will be throwing an error related to initializer):
!python /content/remove_initializer_from_input.py --input /content/deoldify-quant.onnx --output /content/deoldify-quant-clear.onnxTo see quantized DeOldify model at work, visit the following page: Demo