Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

can you release the inference code? #18

Open
zhujunwen opened this issue Jun 25, 2019 · 1 comment
Open

can you release the inference code? #18

zhujunwen opened this issue Jun 25, 2019 · 1 comment

Comments

@zhujunwen
Copy link

can you tell me how to inference the mask with your network?i wrote one,but it‘s result is not good.

@shanyiWang
Copy link

# -*- coding: utf-8 -*-
from __future__ import print_function

import numpy as np
import scipy.misc
import torch
import os

root_dir          = "CamVid/"
label_colors_file = os.path.join(root_dir, "label_colors.txt")
label2color = {}
color2label = {}
label2index = {}
index2label = {}

n_class = 32
means = np.array([103.939, 116.779, 123.68]) / 255.

model_path = "XXXXXXXXXXXXXXXXXXXXX"
model = torch.load(model_path)
use_gpu = torch.cuda.is_available()
if use_gpu:
    model = model.cuda()
model.eval()


def parse_label():
    f = open(label_colors_file, "r").read().split("\n")[:-1]  # ignore the last empty line
    for idx, line in enumerate(f):
        label = line.split()[-1]
        color = tuple([int(x) for x in line.split()[:-1]])
        print(label, color)
        label2color[label] = color
        color2label[color] = label
        label2index[label] = idx
        index2label[idx]   = label


def test_img(img_path):
    img = scipy.misc.imread(img_path, mode='RGB')
    h, w, c = img.shape[0], img.shape[1], img.shape[2]
    val_h = int(h / 32) * 32
    val_w = w
    img = scipy.misc.imresize(img, (val_h, val_w), interp='bilinear', mode=None)

    img = img[:, :, ::-1]
    img = np.transpose(img, (2, 0, 1)) / 255.
    img[0] -= means[0]
    img[1] -= means[1]
    img[2] -= means[2]

    inputs = torch.from_numpy(img.copy()).float()
    inputs = torch.unsqueeze(inputs, 0).cuda()
    output = model(inputs)
    output = output.data.cpu().numpy()

    N, _, h, w = output.shape
    assert (N == 1)
    pred = output.transpose(0, 2, 3, 1).reshape(-1, n_class).argmax(axis=1).reshape(h, w)

    pred_img = np.zeros((val_h, val_w, 3), dtype=np.float32)
    for cls in range(n_class):
        pred_inds = pred == cls
        label = index2label[cls]
        color = label2color[label]
        pred_img[pred_inds] = color
    pred_img = scipy.misc.imresize(pred_img, (h, w), interp='bilinear', mode=None)
    scipy.misc.imsave('result.png', pred_img)


parse_label()
img_path = "XXXXXXXXXXXXXXXXXXXXX"
test_img(img_path)

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

2 participants