前期已经基于OpenVINO搭建成功了天空识别模型，并且能够得到着色的结果图片，下一步就是继续来实现“天空替换”
一、天空替换重构
在OpenVINO着色结果基础上，重新编写c++和python版本的天空替换代码。
// 天空之子算法研究
// 2019年11月30日
#
include
"pch.h"
#
include
"cv_helper.h"
using
namespace cv;
using
namespace std;
int main()
{
Mat matSrc
= imread(
"E:\\未来项目\\天空替换(天空分割)\\测试图片\\测试图片\\sky14.jpg");
Mat matCloud
= imread(
"E:\\未来项目\\你的名字滤镜\\算法实验\\算法实验\\sky1.png");
Mat matMask
= imread(
"E:\\未来项目\\天空替换(天空分割)\\skyInBlue.png",
0);
matMask
= (matMask
==
124);
//统一按照matSrc的大小进行缩放
resize(matCloud, matCloud, matSrc.size());
resize(matMask, matMask, matSrc.size());
Point center
= Point (matMask.cols
/
2, matMask.rows
/
2);
//Mat normal_clone;
//cartoonifyImage(matCloud, matCloud);
Mat normal_clone;
seamlessClone(matCloud, matSrc, matMask, center, normal_clone, MIXED_CLONE);
cv
:
:waitKey();
}
import cv2
import numpy as np
matSrc =cv2.imread( 'E:/template/sky14.jpg')
matCloud = cv2.imread( 'E:/template/cloud3.jpg')
matMask = cv2.imread( 'E:/template/skyInBlue.png', 0)
rows,cols=matMask.shape
for i in range(rows):
for j in range(cols):
if (matMask[i,j]== 124):
matMask[i,j]= 255
else:
matMask[i,j]= 0
height,width=matSrc.shape[: 2]
matCloud=cv2.resize(matCloud,(width,height),interpolation=cv2.INTER_CUBIC)
matMask=cv2.resize(matMask,(width,height),interpolation=cv2.INTER_CUBIC)
center = (width // 2, height // 2)
# Seamlessly clone src into dst and put the results in output
normal_clone = cv2.seamlessClone(matCloud, matSrc, matMask, center, cv2.NORMAL_CLONE)
mixed_clone = cv2.seamlessClone(matCloud, matSrc, matMask, center, cv2.MIXED_CLONE)
cv2.imshow( 'normal_clone',normal_clone)
cv2.imshow( 'mixed_clone',mixed_clone)
cv2.waitKey( 0)
原图：
替换图：
这个已经实现了天空替换的结果，但是颜色还需要调亮一点。
二、 相关的代码融合
view，主要是融入代码：
def process_detail(request,param1):
options = [(
'grpc.max_receive_message_length',
100 *
1024 *
1024),(
'grpc.max_send_message_length',
100 *
1024 *
1024)]
channel = grpc.insecure_channel(
"{}:{}".format(
'localhost',
9000),options = options)
stub = prediction_service_pb2_grpc.PredictionServiceStub(channel)
batch_size =
1
#
TODO
filepath
output_str=
'filepath'
imgfile = os.path.join(
'/root/mysites/goApp/images',param1)
print(imgfile)
img = load_image(imgfile)
imgs = np.zeros((
0,
3,
1024,
2048), np.dtype(
'<f'))
imgs = np.append(imgs, img, axis=
0)
request = predict_pb2.PredictRequest()
request.model_spec.name =
"semantic-segmentation-adas"
print(
"\nRequest shape", img.shape)
img = imgs[
0:
1]
request.inputs[
"data"].CopyFrom(make_tensor_proto(img, shape=(img.shape)))
result = stub.Predict(request,
10.0)
# result includes a dictionary with all model outputs print(img.shape)
output = make_ndarray(result.outputs[
"4455.1"])
for y
in range(
0,img.shape[
0]):
# iterate over responses from all images in the batch
img_out = output[y,:,:,:]
print(
"image in batch item",y,
", output shape",img_out.shape)
img_out = img_out.transpose(
1,
2,
0)
print(
"saving result to",os.path.join(
'/root/mysites/goApp/results',param1+
'.result.jpg'))
out_h, out_w,_ = img_out.shape
print(out_h)
print(out_w)
for batch, data
in enumerate(output):
classes_map = np.zeros(shape=(out_h, out_w,
3), dtype=np.int)
for i
in range(out_h):
for j
in range(out_w):
if len(data[:, i, j]) ==
1:
pixel_class = int(data[:, i, j])
else:
pixel_class = np.argmax(data[:, i, j])
classes_map[i, j, :] = classes_color_map[min(pixel_class,
20)]
classes_map = np.uint8(classes_map)
matMask = cv2.cvtColor(classes_map,cv2.COLOR_BGR2GRAY)
matMask = np.uint8(matMask)
matCloud = cv2.imread(
'/root/mysites/goApp/images/cloud3.jpg')
rows,cols=matMask.shape
for i
in range(rows):
for j
in range(cols):
if (matMask[i,j]==
134):
matMask[i,j]=
255
else:
matMask[i,j]=
0
matsrc = cv2.imread(imgfile)
matsrc = cv2.resize(matsrc,(out_w,out_h),interpolation=cv2.INTER_CUBIC)
matCloud=cv2.resize(matCloud,(out_w,out_h),interpolation=cv2.INTER_CUBIC)
matMask=cv2.resize(matMask,(out_w,out_h),interpolation=cv2.INTER_CUBIC)
center = (out_w //
2, out_h //
2)
normal_clone = cv2.seamlessClone(matCloud,matsrc, matMask, center, cv2.NORMAL_CLONE)
output_str = os.path.join(
'/root/mysites/goApp/results',param1+
'.result.jpg')
cv2.imwrite(output_str,normal_clone)
return HttpResponse(output_str)
最后实现在浏览器中的调用时正常的。
三、 目标导向
我最终想实现的是完全自可 主 控的类似https://cloud.baidu.com/product/imageprocess/sky_seg 的服务
包括网站服务，后端调用等。当然这个界面比较复杂，我自己的实现比较简单，如果能够找到 Django的模板的话，我也会来进行实现。