predictMulti.py

import argparse
import logging
import os

import numpy as np
import torch
import torch.nn.functional as F
from PIL import Image
from torchvision import transforms

import sys
sys.path.append('../pytorch-unet')

from unet import UNet
from data_vis import plot_img_and_mask
from dataset import BasicDataset

# python3 predictMulti.py --model checkpoints_WF_adult_large_small_with_aphids_warmRestarts/CP_epoch25.pth  --input /home/rob/Pictures/LesBargefieldstation/FullSize/*.jpg --scale 1 --mask-threshold 0.65 --path /home/rob/Pictures/LesBargefieldstation/FullSize/

def get_mask(net_output, resize) :
    probs = net_output  # we removed the extra sigmoid

    probs = probs.squeeze(0)


    tf = transforms.Compose(
        [
            transforms.ToPILImage(),
            transforms.Resize(resize),
            transforms.ToTensor()
        ]
    )

    probs = tf(probs.cpu())
    full_mask = probs.squeeze().cpu().numpy()

    return full_mask

def predict_img(net,
                full_img,
                device,
                scale_factor=1,
                out_threshold=0.5):
    net.eval()

    img = torch.from_numpy(BasicDataset.preprocess(full_img, scale_factor))

    img = img.unsqueeze(0)
    img = img.to(device=device, dtype=torch.float32)

    img_size = img.size()
    print(img_size)
    crop_number = 10
    img_size = (img_size[2], img_size[3]) # height and width as tuple
    if img_size[0]>2048 and img_size[1]>2048 :
        full_mask = torch.zeros(img.size())
        crop_x = img_size[0]//crop_number # a double // converts to an int from the double
        print(img_size)
        crop_y = img_size[1]//crop_number
        with torch.no_grad():
            for x in range(crop_number) :
                for y in range(crop_number) :
                    #print(str(crop_x))
                    full_mask[:,:,x*crop_x : x*crop_x+crop_x, y*crop_y : y*crop_y+crop_y] = net(img[:,:,x*crop_x : x*crop_x+crop_x, y*crop_y : y*crop_y+crop_y] )
    else :
        full_mask = net(img)

    full_mask= get_mask(full_mask, full_img.size[1])

    print(full_mask.max())
    return full_mask


def get_args():
    parser = argparse.ArgumentParser(description='Predict masks from input images',
                                     formatter_class=argparse.ArgumentDefaultsHelpFormatter)
    parser.add_argument('--model', '-m', default='MODEL.pth',
                        metavar='FILE',
                        help="Specify the file in which the model is stored")
    parser.add_argument('--input', '-i', metavar='INPUT', nargs='+',
                        help='filenames of input images', required=True)
    parser.add_argument('--path', '-p', metavar='INPUT', nargs='+',
                        help='Path of ouput images')  # RJL put this in to specify path when outputting multiple files so not hard wired in code
    parser.add_argument('--output', '-o', metavar='INPUT', nargs='+',
                        help='Filenames of ouput images')
    parser.add_argument('--viz', '-v', action='store_true',
                        help="Visualize the images as they are processed",
                        default=False)
    parser.add_argument('--no-save', '-n', action='store_true',
                        help="Do not save the output masks",
                        default=False)
    parser.add_argument('--mask-threshold', '-t', type=float,
                        help="Minimum probability value to consider a mask pixel white",
                        default=0.5)
    parser.add_argument('--scale', '-s', type=float,
                        help="Scale factor for the input images",
                        default=0.5)

    return parser.parse_args()


def get_output_filenames(args):
    in_files = args.input
    output_path = args.path[0] # RJL read the path in from args
    out_files = []

    if not args.output:
        for f in in_files:
            #pathsplit = os.path.splitext(f)
            fname = os.path.basename(f)
            #out_files.append("{}_OUT{}".format(pathsplit[0], pathsplit[1]))
            out_files.append("{}_OUT{}".format(output_path, fname))  # outputs all files to a set folder
    elif len(in_files) != len(args.output):
        logging.error("Input files and output files are not of the same length")
        raise SystemExit()
    else:
        out_files = args.output

    return out_files


def mask_to_image(mask):
    return Image.fromarray((mask * 255).astype(np.uint8))


if __name__ == "__main__":
    args = get_args()
    in_files = args.input
    out_files = get_output_filenames(args)

    net = UNet(n_channels=3, n_classes=3)

    logging.info("Loading model {}".format(args.model))

    device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
    logging.info(f'Using device {device}')
    net.to(device=device)
    net.load_state_dict(torch.load(args.model, map_location=device))

    logging.info("Model loaded !")

    for i, fn in enumerate(in_files): # can input multiple files
        logging.info("\nPredicting image {} ...".format(fn))

        img = Image.open(fn)

        mask = predict_img(net=net,
                           full_img=img,
                           scale_factor=args.scale,
                           out_threshold=args.mask_threshold,
                           device=device)

        if not args.no_save:
            out_fn = out_files[i]

            # get mask dimensions for rgb output image
            arrayWidth = mask.shape[1]
            arrayHeight = mask.shape[2]

            # make output rgb to be filled with mask channels
            result_rgb = Image.new("RGB", [arrayWidth, arrayHeight], 255)
            
            result_r = mask_to_image(mask[0,:,:])
            result_g = mask_to_image(mask[1,:,:])
            result_b = mask_to_image(mask[2,:,:])
            result_r.save(out_files[i] + '0.png')
            result_g.save(out_files[i] + '1.png')
            result_b.save(out_files[i] + '2.png')

            # use the split and merge commands to unpack and pack rgb images
            result_rgb = Image.merge('RGB', (result_r, result_g, result_b))

            result_rgb.save(out_files[i] + 'ALL.png') # save out rgb mask

            logging.info("Mask saved to {}".format(out_files[i]))

        if args.viz:
            logging.info("Visualizing results for image {}, close to continue ...".format(fn))
            plot_img_and_mask(img, mask)