DeepLearning/Image_segmentation/DeepLabV3/predict.py at master · Overmind7/DeepLearning · GitHub

History

172 lines (145 loc) · 6.42 KB

Raw

100

101

102

103

104

105

106

107

108

109

110

111

112

113

114

115

116

117

118

119

120

121

122

123

124

125

126

127

128

129

130

131

132

133

134

135

136

137

138

139

140

141

142

143

144

145

146

147

148

149

150

151

152

153

154

155

156

157

158

159

160

161

162

163

164

165

166

167

168

169

170

171

172

import glob

import json

import logging

import os.path

import time

import timeit

from pathlib import Path

import numpy as np

import argparse

import yaml

import shutil

from tqdm import tqdm

from PIL import Image

import matplotlib.pyplot as plt

import torch

import torch.nn.functional as F

from torch.utils.data import DataLoader

from torchvision import transforms

from models.network import build_model

from utils.label import label2rgb

IMG_FORMATS = ['bmp', 'jpg', 'jpeg', 'png', 'tif', 'tiff', 'dng', 'webp', 'mpo']

def time_sync():

# pytorch-accurate time

if torch.cuda.is_available():

torch.cuda.synchronize()

return time.time()

@torch.no_grad()

def run(

cfg='config/example.yaml', # 配置文件，主要用于读取模型配置

weights='best_model.pth', # 模型路径

source='./data/test', # 测试数据路径，可以是文件夹，可以是单张图片

use_cuda=True, # 是否使用cuda

view_img=False, # 是否可视化测试图片

save_mask=True, # 是否将保存mask

save_viz=True, # 是否保存融合图

palette_path="palette.json",

project='result' # 结果输出路径

if isinstance(cfg, dict):

config = cfg

else:

import yaml

yaml_file = Path(cfg).name

with open(cfg) as f:

config = yaml.safe_load(f)

device = torch.device("cuda" if torch.cuda.is_available() and use_cuda else "cpu")

if save_mask:

os.makedirs(project + "/mask", exist_ok=True)

if save_viz:

os.makedirs(project + "/viz", exist_ok=True)

# Load model

assert os.path.exists(weights), "model path: {} does not exists".format(weights)

num_classes = config["train"]["num_classes"] + 1

model = build_model(config, num_classes, pretrain=False)

checkpoint = torch.load(weights, map_location='cpu')["model"]

model.load_state_dict(checkpoint, strict=True)

model.eval().to(device)

# 调色板，用于给mask上色

with open(palette_path, "rb") as f:

pallette_dict = json.load(f)

pallette = []

for v in pallette_dict.values():

pallette += v

# Run once

y = model(torch.rand(1, 3, 480, 480).to(device))

# 转成torch script

# 如果libtorch版本太低的话，模型版本也不能太高，需要进行转换

# checkpoint = torch.load(weights, map_location='cpu')["model"]

# torch.save(checkpoint, "libtorch.pth", _use_new_zipfile_serialization=False)

# trace_module = torch.jit.trace(model, torch.rand(1, 3, 224, 224).to(device))

# print(trace_module.code) # 查看模型结构

# output = trace_module(torch.ones(1, 3, 224, 224).to(device)) # 测试

# print(output)

# trace_module.save('model_gpu.pt') # 模型保存

# script_module = torch.jit.script(model)

# print(script_module.code)

# # output = script_module(torch.rand(1, 1, 224, 224))

# script_module.save('model.pt')

# Data transform

data_transform = transforms.Compose([transforms.Resize(1024),

transforms.ToTensor(),

transforms.Normalize(mean=(0.485, 0.456, 0.406),

std=(0.229, 0.224, 0.225))

])

# Load img

assert os.path.exists(source), "data source: {} does not exists".format(source)

if os.path.isdir(source):

files = sorted(glob.glob(os.path.join(source, '*.*')))

elif os.path.isfile(source):

# img = Image.open(source)

# if img.mode != 'RGB':

# img = img.convert('RGB')

files = [source]

else:

raise Exception(f'ERROR: {source} does not exist')

images = [x for x in files if x.split('.')[-1].lower() in IMG_FORMATS]

images = tqdm(images)

for img_path in images:

full_img = Image.open(img_path).convert("RGB")

img = data_transform(full_img)

img = torch.unsqueeze(img, dim=0)

t1 = time_sync()

output = model(img.to(device))

t2 = time_sync()

print("inference time: {}".format(t2 - t1))

prediction = output['out'].argmax(1).squeeze(0)

prediction = prediction.to("cpu").numpy().astype(np.uint8)

file_name = img_path.split(os.sep)[-1][:-4]

mask = Image.fromarray(prediction)

mask.putpalette(pallette)

mask = mask.resize((full_img.size[0], full_img.size[1]), resample=Image.NEAREST)

viz = label2rgb(np.asarray(mask), np.asarray(full_img), font_size=15, loc="rb",

colormap=np.asarray(list(pallette_dict.values())))

if view_img:

fig, ax = plt.subplots(1, 3)

ax[0].set_title('Input image')

ax[0].imshow(full_img)

ax[1].set_title(f'Output mask')

ax[1].imshow(mask)

ax[2].set_title(f'Vis')

ax[2].imshow(viz)

plt.xticks([]), plt.yticks([])

plt.show()

if save_mask:

mask.save(os.path.join(project + "/mask", "{}.png".format(file_name)))

if save_viz:

viz = Image.fromarray(viz)

viz.save(os.path.join(project + "/viz", "{}.png".format(file_name)))

if __name__ == '__main__':

parser = argparse.ArgumentParser(description='Predict masks from input images')

parser.add_argument('--cfg', type=str, default='config/example.yaml',

help='experiment configure file name')

parser.add_argument('--weights', '-w', default='best.pth', metavar='FILE',

help='Specify the file in which the model is stored')

parser.add_argument('--source', '-i', default='data/test',

help='Filenames of input images/dir')

parser.add_argument('--use_cuda', default=True, action='store_true', help='Use cuda to predict')

parser.add_argument('--view_img', '-v', action='store_true', default=False,

help='Visualize the images as they are processed')

parser.add_argument('--save-mask', '-sm', action='store_true', default=True, help='Whether save the output masks')

parser.add_argument('--save-viz', '-sv', action='store_true', default=True, help='Whether save the output vizs')

parser.add_argument('--palette_path', default='palette.json', help='Plot Mask by different color')

parser.add_argument('--project', '-p', metavar='OUTPUT', default='result', help='Output of img path')

opt = parser.parse_args()

run(**vars(opt))

DeepLearning/Image_segmentation/DeepLabV3/predict.py at master · Overmind7/DeepLearning · GitHub

Expand file tree

Breadcrumbs

predict.py

Latest commit

History

Breadcrumbs

predict.py

File metadata and controls