live2diff/MiDaS/tf/run_onnx.py

"""Compute depth maps for images in the input folder.
"""
import os
import glob
import utils
import cv2
import sys
import numpy as np
import argparse

import onnx
import onnxruntime as rt

from transforms import Resize, NormalizeImage, PrepareForNet


def run(input_path, output_path, model_path, model_type="large"):
    """Run MonoDepthNN to compute depth maps.

    Args:
        input_path (str): path to input folder
        output_path (str): path to output folder
        model_path (str): path to saved model
    """
    print("initialize")

    # select device
    device = "CUDA:0"
    #device = "CPU"
    print("device: %s" % device)

    # network resolution
    if model_type == "large":
        net_w, net_h = 384, 384
    elif model_type == "small":
        net_w, net_h = 256, 256
    else:
        print(f"model_type '{model_type}' not implemented, use: --model_type large")
        assert False

    # load network
    print("loading model...")
    model = rt.InferenceSession(model_path)
    input_name = model.get_inputs()[0].name
    output_name = model.get_outputs()[0].name

    resize_image = Resize(
                net_w,
                net_h,
                resize_target=None,
                keep_aspect_ratio=False,
                ensure_multiple_of=32,
                resize_method="upper_bound",
                image_interpolation_method=cv2.INTER_CUBIC,
            )
    
    def compose2(f1, f2):
        return lambda x: f2(f1(x))

    transform = compose2(resize_image, PrepareForNet())

    # get input
    img_names = glob.glob(os.path.join(input_path, "*"))
    num_images = len(img_names)

    # create output folder
    os.makedirs(output_path, exist_ok=True)

    print("start processing")

    for ind, img_name in enumerate(img_names):

        print("  processing {} ({}/{})".format(img_name, ind + 1, num_images))

        # input
        img = utils.read_image(img_name)
        img_input = transform({"image": img})["image"]

        # compute
        output = model.run([output_name], {input_name: img_input.reshape(1, 3, net_h, net_w).astype(np.float32)})[0]
        prediction = np.array(output).reshape(net_h, net_w)
        prediction = cv2.resize(prediction, (img.shape[1], img.shape[0]), interpolation=cv2.INTER_CUBIC)
       
        # output
        filename = os.path.join(
            output_path, os.path.splitext(os.path.basename(img_name))[0]
        )
        utils.write_depth(filename, prediction, bits=2)

    print("finished")


if __name__ == "__main__":
    parser = argparse.ArgumentParser()

    parser.add_argument('-i', '--input_path', 
        default='input',
        help='folder with input images'
    )

    parser.add_argument('-o', '--output_path', 
        default='output',
        help='folder for output images'
    )

    parser.add_argument('-m', '--model_weights', 
        default='model-f6b98070.onnx',
        help='path to the trained weights of model'
    )

    parser.add_argument('-t', '--model_type', 
        default='large',
        help='model type: large or small'
    )

    args = parser.parse_args()

    # compute depth maps
    run(args.input_path, args.output_path, args.model_weights, args.model_type)