1、model
模型选择为unet,数据集为自己训练的剥落、钢筋外露、裂缝三类混凝土缺陷语义分割检测。
import torch.nn as nn
import torch
import torch.nn.functional as F
import time
class DoubleConv(nn.Module):
def __init__(self, in_channels, out_channels, mid_channels=None):
super().__init__()
if not mid_channels:
mid_channels = out_channels
self.double_conv = nn.Sequential(
nn.Conv2d(in_channels, mid_channels, kernel_size=3, padding=1, bias=False),
nn.BatchNorm2d(mid_channels),
nn.ReLU(inplace=True),
nn.Conv2d(mid_channels, out_channels, kernel_size=3, padding=1, bias=False),
nn.BatchNorm2d(out_channels),
nn.ReLU(inplace=True))
def forward(self, x):
return self.double_conv(x)
class Down(nn.Module):
def __init__(self, in_channels, out_channels):
super().__init__()
self.maxpool_conv = nn.Sequential(
nn.MaxPool2d(2),
DoubleConv(in_channels, out_channels))
def forward(self, x):
return self.maxpool_conv(x)
class Up(nn.Module):
def __init__(self, in_channels, out_channels, bilinear=True):
super().__init__()
if bilinear:
self.up = nn.Upsample(scale_factor=2, mode='bilinear', align_corners=True)
self.conv = DoubleConv(in_channels, out_channels, in_channels // 2)
else:
self.up = nn.ConvTranspose2d(in_channels, in_channels // 2, kernel_size=2, stride=2)
self.conv = DoubleConv(in_channels, out_channels)
def forward(self, x1, x2):
x1 = self.up(x1)
# input is CHW
diffY = x2.size()[2] - x1.size()[2]
diffX = x2.size()[3] - x1.size()[3]
x1 = F.pad(x1, [diffX // 2, diffX - diffX // 2,
diffY // 2, diffY - diffY // 2])
x = torch.cat([x2, x1], dim=1)
return self.conv(x)
class OutConv(nn.Module):
def __init__(self, in_channels, out_channels):
super(OutConv, self).__init__()
self.conv = nn.Conv2d(in_channels, out_channels, kernel_size=1)
def forward(self, x):
return self.conv(x)
class unet(nn.Module):
def __init__(self, n_classes, bilinear=True):
super(unet, self).__init__()
self.n_channels = 3
self.n_classes = n_classes
self.bilinear = bilinear
self.inc = DoubleConv(self.n_channels, 64)
self.down1 = Down(64, 128)
self.down2 = Down(128, 256)
self.down3 = Down(256, 512)
factor = 2 if bilinear else 1
self.down4 = Down(512, 1024 // factor)
self.up1 = Up(1024, 512 // factor, bilinear)
self.up2 = Up(512, 256 // factor, bilinear)
self.up3 = Up(256, 128 // factor, bilinear)
self.up4 = Up(128, 64, bilinear)
self.outc = OutConv(64, n_classes)
def forward(self, x):
x1 = self.inc(x)
x2 = self.down1(x1)
x3 = self.down2(x2)
x4 = self.down3(x3)
x5 = self.down4(x4)
x = self.up1(x5, x4)
x = self.up2(x, x3)
x = self.up3(x, x2)
x = self.up4(x, x1)
logits = self.outc(x)
return logits
2、使用pytorch-gpu推理
def pytorch_gpu():
model_path = r"F:\Learning_materials\learn_cpp\project\crack_detection\model\best_epoch_weights.pth"
num_classes = 4
name_classes = ["background","spall", "crack", "rebar"]#/opt/data/private/muti_damage_detection/muti_seg/image/train.txt
gt_dir = "/opt/data/private/muti_damage_detection/seg/dataset/"
pred_dir = os.path.join(r"F:\Learning_materials\learn_cpp\project\crack_detection\model\result", 'detection-results')
rgb_dir = os.path.join(r"F:\Learning_materials\learn_cpp\project\crack_detection\model\result", 'detection_predict')
# image_ids = open(os.path.join("/opt/data/private/muti_damage_detection/muti_seg/dataset/","val.txt"),'r').read().splitlines()
image_ids = glob.glob(r"F:\Learning_materials\learn_cpp\project\crack_detection\img\*.jpg")
if not os.path.exists(pred_dir):
os.makedirs(pred_dir)
if not os.path.exists(rgb_dir):
os.makedirs(rgb_dir)
model = unet(4)#torch.cuda.is_available()
net = model.eval()
device = torch.device('cuda' if False else 'cpu')
net.load_state_dict(torch.load(model_path, map_location=device))
net = net.to(device)
print('Load model.\n{} model, and classes loaded.'.format(model_path))
totol_time = 0
for image_id in tqdm(image_ids):
image_path = image_id.split(",")[0]
image_id2 = image_path.split("\\")[-1].split(".")[0]
image = Image.open(image_path)
width, height = image.size
target_size = max(width, height)
resized_image = image.resize((target_size, target_size))
cropped_image = resized_image.crop(((target_size - width) // 2, (target_size - height) // 2, (target_size + width) // 2, (target_size + height) // 2))
image = image.resize((256, 256))
image,inference_time = get_miou_png(image,net,device)
image.save(os.path.join(pred_dir, image_id2 + ".png"))
image_data = np.array(image)
image_data = out_to_rgb_np(image_data)
cv2.imwrite(os.path.join(rgb_dir, image_id2 + ".png"), image_data)
totol_time += inference_time
print("mainInference time:", (totol_time)/10, "seconds")
3、使用onnxruntime-gpu推理
def onnxruntime_gpu():
ort_session = ort.InferenceSession('F:\Learning_materials\learn_cpp\project\crack_detection\model\model_softmax_argmax2.onnx',providers=["CUDAExecutionProvider"])
image_ids = glob.glob(r"F:\Learning_materials\learn_cpp\project\crack_detection\img\*.jpg")
totol_time = 0
for i in range(10):
input_shape = (1, 3, 256, 256) # 定义输入形状torch.cuda.is_available()
input_data = np.random.rand(*input_shape).astype(np.float32)
start_time = time.time()
output = ort_session.run(None, {'input': input_data})
end_time = time.time()
inference_time = end_time - start_time
totol_time += (end_time - start_time)
print('Inference time for', ':', inference_time, 'seconds')
print("mainInference time:", (totol_time)/10, "seconds")
安装并测试onnxruntime
pip install onnxruntime
pip install onnxruntime-gpu