def test6():
# #读取图像到数组中
t = util.imageToTensor(
"/home/xws/Downloads/300w_cropped/01_Indoor/indoor_001.png")
img = util.tensorToImage(t[0])
width = t[1]
height = t[2]
m = matrix.Matrix(height, width)
m_point = matrix.Matrix(height, width)
path = "/home/xws/Downloads/300w_cropped/01_Indoor/indoor_001.png"
img = Image.open(path)
# imgTensor.type(torch.DoubleTensor)
scale = 0.6
angle = 30
translation = 0.1
new_img = tfs.functional.rotate(img, -angle)
new_img = pic_strong(img).type(torch.DoubleTensor)
img_torch = new_img
# m.scale(1 / scale, 1 / scale)
m.translation(-translation, -translation)
# m_point.scale(scale, scale)
# m_point.rotation(angle)
m_point.translation_point(translation, translation)
# theta = torch.tensor([[1, 0, 0.2], [0, 1, -0.4]], dtype=torch.double)
theta = torch.from_numpy(m.to_theta())
grid = F.affine_grid(theta.unsqueeze(0),
img_torch.unsqueeze(0).size(), False)
output = F.grid_sample(img_torch.unsqueeze(0), grid)
new_img_torch = output[0]
# plt.imshow(new_img_torch.numpy().transpose(1, 2, 0))
img = util.tensorToImage(output)
plt.imshow(img)
points = util.textToPoint(
"/home/xws/Downloads/300w_cropped/01_Indoor/indoor_001.pts")
points = m_point.dot_point_68(points)[1]
# 使用红色星状物标记绘制点
i = 1
for p in points:
plt.plot(p[0], p[1], "r_")
i = i + 1
plt.show()
print("this is end")
def show(plt, X, L):
plt.cla()
img = util.tensorToImage(X)
plt.imshow(img)
points = util.tensorToPoint(L.cpu().detach())
for p in points:
plt.plot(p[0] * Config.IMAGE_SIZE, p[1] * Config.IMAGE_SIZE, "r.")
plt.savefig("/home/xws/Downloads/python/python/face/img/test0.png")
def show(model, data, path):
imgTensor = data[0]
X = imgTensor.view(1, 3, Config.IMAGE_SIZE, Config.IMAGE_SIZE)
testout = model(X)
points = util.tensorToPoint(testout.cpu().detach())
for p in points:
plt.plot(p[0] * Config.IMAGE_SIZE, p[1] * Config.IMAGE_SIZE, "r+")
img = util.tensorToImage(imgTensor)
plt.imshow(img)
plt.savefig(path[0].replace("300w_cropped/01_Indoor", "test"))
# plt.show()
plt.cla()
def test3():
# #读取图像到数组中
# im = array(Image.open('/home/xws/Downloads/boll.jpeg'))
t = util.imageToTensor(
"/home/xws/Downloads/300w_cropped/01_Indoor/indoor_001.png")
img = util.tensorToImage(t[0])
# img = plt.imread("/home/xws/Downloads/300w_cropped/01_Indoor/indoor_300.png")
width = t[1]
height = t[2]
m = matrix.Matrix(height, width)
m.rotation(30)
path = "/home/xws/Downloads/300w_cropped/01_Indoor/indoor_001.png"
img = Image.open(path)
new_img = transform(img)
new_img = tfs.functional.rotate(img, -30)
plt.imshow(new_img)
# text = util.readText("/home/xws/Downloads/300w_cropped/01_Indoor/indoor_300.pts")
points = util.textToPoint(
"/home/xws/Downloads/300w_cropped/01_Indoor/indoor_001.pts")
points = m.dot_point_68(points)[1]
# # 使用红色星状物标记绘制点
i = 1
for p in points:
plt.plot(p[0], p[1], "r_")
# plt.text(p[0], p[1], i)
i = i + 1
# plt.text(points[8][0], points[8][1], 0)
# plt.text(points[17][0], points[17][1], 0)
# plt.text(points[26][0], points[26][1], 0)
# plt.text(points[27][0], points[27][1], 0)
# plt.text(points[33][0], points[33][1], 0)
# plt.text(points[48][0], points[48][1], 0)
# plt.text(points[54][0], points[54][1], 0)
# plt.plot(points[8][0], points[8][1], "g|")
plt.plot(points[17][0], points[17][1], "g|")
plt.plot(points[26][0], points[26][1], "g|")
# plt.plot(points[27][0], points[27][1], "g|")
plt.plot(points[33][0], points[33][1], "g|")
plt.plot(points[48][0], points[48][1], "g|")
plt.plot(points[54][0], points[54][1], "g|")
# #添加标题，显示绘制的图像
plt.show()
print("this is end")
def test4():
# #读取图像到数组中
t = util.imageToTensor(
"/home/xws/Downloads/300w_cropped/01_Indoor/indoor_001.png")
img = util.tensorToImage(t[0])
width = t[1]
height = t[2]
m = matrix.Matrix(height, width)
path = "/home/xws/Downloads/300w_cropped/01_Indoor/indoor_001.png"
img = Image.open(path)
# new_img = transform(img)
# new_img = tfs.functional.rotate(img, -30)
new_img = tfs.functional.resize(img,
[Config.IMAGE_SIZE, Config.IMAGE_SIZE],
Image.BICUBIC)
plt.imshow(img)
plt.imshow(new_img)
points = util.textToPoint(
"/home/xws/Downloads/300w_cropped/01_Indoor/indoor_001.pts")
# 将point扩充一下
newpoints = np.arange(68 * 3, dtype=float).reshape(68, 3)
for i, p in enumerate(points):
print("ww", i, p)
newpoints[i][0] = p[0]
newpoints[i][1] = p[1]
newpoints[i][2] = 1
print(newpoints)
m.rotation(30)
# m.scale(1, 1)
pp = m.dot(newpoints)
for i, p in enumerate(pp):
print("ww", i, p)
points[i][0] = p[0]
points[i][1] = p[1]
# 使用红色星状物标记绘制点
i = 1
for p in points:
plt.plot(p[0], p[1], "r_")
i = i + 1
plt.show()
print("this is end")
def test0():
plt.ion()
# #读取图像到数组中
# im = array(Image.open('/home/xws/Downloads/boll.jpeg'))
f = plt.figure()
# #绘制图像
t = util.imageToTensor(
"/home/xws/Downloads/300w_cropped/01_Indoor/indoor_001.png")
img = util.tensorToImage(t[0])
# img = plt.imread("/home/xws/Downloads/300w_cropped/01_Indoor/indoor_300.png")
plt.imshow(img)
width = t[1]
height = t[2]
# text = util.readText("/home/xws/Downloads/300w_cropped/01_Indoor/indoor_300.pts")
points = util.textToPoint(
"/home/xws/Downloads/300w_cropped/01_Indoor/indoor_001.pts")
# print("points = ", points)
points = util.pointToTensor(points)
# print("pointToTensor = ", points)
points = util.tensorToPoint(points)
# print("tensorToPoint = ", points)
# # 使用红色星状物标记绘制点
for p in points:
plt.plot(p[0], p[1], "r+")
# #绘制前两个点的线
# plt.plot(x[:2], y[:2])
# #添加标题，显示绘制的图像
plt.title('Plotting:"pic1.png"')
plt.ioff()
plt.savefig("/home/xws/Downloads/python/python/face/test.png")
# plt.show()
print("this is end")
# 加载小批次数据，即将MNIST数据集中的data分成每组batch_size的小块，shuffle指定是否随机读取
# train_loader = Data.DataLoader(
# dataset=dataset.IBUGDataSet(Config.BATCH_SIZE),
# batch_size=Config.BATCH_SIZE,
# shuffle=True,
# )
# load test
testTensor = util.imageToTensor("/home/xws/Downloads/camera/25.jpeg")
# testTensor = util.imageToTensor(
# "/home/xws/Downloads/300w_cropped/01_Indoor/indoor_001.png"
# )
imgTensor = testTensor[0]
img = util.tensorToImage(testTensor[0])
plt.imshow(img)
def test():
model.eval() # 模型转化为评估模式
if device == "cpu":
X = imgTensor # 包装tensor用于自动求梯度
else:
X = imgTensor.cuda() # 包装tensor用于自动求梯度
X = X.view(1, 3, Config.IMAGE_SIZE, Config.IMAGE_SIZE)
testout = model(X)
show(plt, X, testout)
def test7():
# #读取图像到数组中
t = util.imageToTensor(
"/home/xws/Downloads/300w_cropped/01_Indoor/indoor_001.png")
img = util.tensorToImage(t[0])
width = t[1]
height = t[2]
m = matrix.Matrix(height, width)
m_point = matrix.Matrix(height, width)
path = "/home/xws/Downloads/300w_cropped/01_Indoor/indoor_001.png"
img = Image.open(path)
# imgTensor.type(torch.DoubleTensor)
scale = 0.6
angle = 30
translation = 0.1
new_img = tfs.functional.rotate(img, -angle)
new_img = pic_strong(img).type(torch.DoubleTensor)
img_torch = new_img
m.scale(1 / scale, 1 / scale)
m.translation(-translation, -translation)
m_point.scale(scale, scale)
# m_point.rotation(angle)
m_point.translation_point(translation, translation)
# theta = torch.tensor([[1, 0, 0.2], [0, 1, -0.4]], dtype=torch.double)
theta = torch.from_numpy(m.to_theta())
grid = F.affine_grid(theta.unsqueeze(0),
img_torch.unsqueeze(0).size(), False)
output = F.grid_sample(img_torch.unsqueeze(0), grid)
new_img_torch = output[0]
plt.imshow(new_img_torch.numpy().transpose(1, 2, 0))
points = util.textToPoint(
"/home/xws/Downloads/300w_cropped/01_Indoor/indoor_001.pts")
point_tensor = util.pointToTensor(points)
point_tensor = point_tensor.view(1, 68, 2)
grid = F.affine_grid(theta.unsqueeze(0),
point_tensor.unsqueeze(0).size(), False)
output = F.grid_sample(point_tensor.unsqueeze(0), grid)
points = util.tensorToPoint(output[0])
# 将point扩充一下
newpoints = np.arange(68 * 3, dtype=float).reshape(68, 3)
for i, p in enumerate(points):
newpoints[i][0] = p[0]
newpoints[i][1] = p[1]
newpoints[i][2] = 1
# m.scale(1, 1)
pp = m_point.dot(newpoints)
for i, p in enumerate(pp):
points[i][0] = p[0]
points[i][1] = p[1]
# 使用红色星状物标记绘制点
i = 1
for p in points:
plt.plot(p[0], p[1], "r_")
i = i + 1
plt.show()
print("this is end")
# img_ = np.rot90(img_, -1).copy()
# if len(face) >= 1:
# (x, y, w, h) = face[0]
# cv2.rectangle(frame, (x, y), (x + w, y + h), (255, 0, 0), 2)
# img = frame[:][y : y + h, x : x + w]
# 如果分类器能捕捉到人脸，就对其进行剪裁送入网络，否则就将整张图片送入
# img = img.astype(np.uint8)
image = Image.fromarray(img)
# plt.imshow(image)
# plt.savefig("/home/xws/Downloads/camera/" + str(camera_index))
# plt.show()
# plt.cla()
camera_index = camera_index + 1
tensor = transforms(image)
img = util.tensorToImage(tensor)
plt.imshow(img)
plt.savefig("/home/xws/Downloads/camera/" + str(camera_index) + ".jpeg")
plt.show()
plt.cla()
tensor1 = tensor.reshape(1, 3, Config.IMAGE_SIZE, Config.IMAGE_SIZE)
pre = model(tensor1)
# util.show(plt,tensor,pre)
points = util.tensorToPoint(pre.detach())
for p in points:
p[0] = p[0] * 480
p[1] = p[1] * 640
cv2.circle(img_, (int(p[0]), int(p[1])), point_size, point_color,
thickness)
# cv2.rectangle(img_, (int(-10), int(0)), (int(200), int(300)), (0, 255, 0))