def bbox_iou2(box1, box2, x1y1x2y2=True):
"""
Returns the IoU of two bounding boxes
"""
if not x1y1x2y2:
# Transform from center and width to exact coordinates
b1_x1, b1_x2 = box1[:, 0] - box1[:, 2] / 2, box1[:, 0] + box1[:, 2] / 2
b1_y1, b1_y2 = box1[:, 1] - box1[:, 3] / 2, box1[:, 1] + box1[:, 3] / 2
b2_x1, b2_x2 = box2[:, 0] - box2[:, 2] / 2, box2[:, 0] + box2[:, 2] / 2
b2_y1, b2_y2 = box2[:, 1] - box2[:, 3] / 2, box2[:, 1] + box2[:, 3] / 2
else:
# Get the coordinates of bounding boxes
b1_x1, b1_y1, b1_x2, b1_y2 = box1[:, 0], box1[:, 1], box1[:,
2], box1[:,
3]
b2_x1, b2_y1, b2_x2, b2_y2 = box2[:, 0], box2[:, 1], box2[:,
2], box2[:,
3]
# get the corrdinates of the intersection rectangle
inter_rect_x1 = np.max(b1_x1, b2_x1)
inter_rect_y1 = np.max(b1_y1, b2_y1)
inter_rect_x2 = np.min(b1_x2, b2_x2)
inter_rect_y2 = np.min(b1_y2, b2_y2)
# Intersection area
inter_area = np.clamp(inter_rect_x2 - inter_rect_x1 + 1, min=0) * np.clamp(
inter_rect_y2 - inter_rect_y1 + 1, min=0)
# Union Area
b1_area = (b1_x2 - b1_x1 + 1) * (b1_y2 - b1_y1 + 1)
b2_area = (b2_x2 - b2_x1 + 1) * (b2_y2 - b2_y1 + 1)
iou = inter_area / (b1_area + b2_area - inter_area + 1e-16)
return iou
def update(self, w, dw, config):
if config['m'] is None: config['m'] = np.zeros_like(w)
if config['v'] is None: config['v'] = np.zeros_like(w)
if config['max_v'] is None: config['max_v'] = np.zeros_like(w)
max_v = config['max_v']
beta1, beta2 = config['beta1'], config['beta2']
t, m, v = config['t'], config['m'], config['v']
t += 1
m = m * beta1 + (1 - beta1) * dw
v = v * beta2 + (1 - beta2) * (dw**2)
if config['amsbound']:
max_v = np.max(max_v, v)
denom = np.sqrt(max_v) + config['epsilon']
else:
denom = np.sqrt(v)
bias_c1 = 1 - beta1**t
bias_c2 = 1 - beta2**t
step_size = config['learning_rate'] * np.sqrt(bias_c2) / bias_c1
lb = config['final_lr'] * (1 - 1 / (config['gamma'] * t + 1))
ub = config['final_lr'] * (1 + 1 / (config['gamma'] * t))
step_size = np.full_like(denom, step_size)
step_size = np.clamp(step_size / denom, lb, ub) * m
w -= step_size
config['m'] = m
config['v'] = v
config['t'] = t
config['max_v'] = max_v
return w, config
def check(self, fore, back):
id = self.slice[0]
template = self.template[:fore.shape[0], :fore.shape[1], :]
fore_warp = torch.round(template + fore).long()
fore_warp[:, :, 0] = torch.clamp(fore_warp[:, :, 0],
min=0,
max=fore.shape[0] - 1)
fore_warp[:, :, 1] = torch.clamp(fore_warp[:, :, 1],
min=0,
max=fore.shape[1] - 1)
fore_result=torch.where(torch.norm(back[fore_warp[:,:,0],fore_warp[:,:,1]]+fore,2,dim=2)<=1, \
self.one,self.zero)
back_warp = torch.round(template + back).long()
back_warp[:, :, 0] = torch.clamp(back_warp[:, :, 0],
min=0,
max=fore.shape[0] - 1)
back_warp[:, :, 1] = torch.clamp(back_warp[:, :, 1],
min=0,
max=fore.shape[1] - 1)
back_result=torch.where(torch.norm(fore[back_warp[:,:,0],back_warp[:,:,1]]+back,2,dim=2)<=1, \
self.one,self.zero)
fore *= fore_result.unsqueeze(-1)
back *= back_result.unsqueeze(-1)
fore_warp = torch.round(template + fore).long()
fore_warp[:, :, 0] = torch.clamp(fore_warp[:, :, 0],
min=0,
max=fore.shape[0] - 1)
fore_warp[:, :, 1] = torch.clamp(fore_warp[:, :, 1],
min=0,
max=fore.shape[1] - 1)
fore_result=torch.where(torch.norm(back[fore_warp[:,:,0],fore_warp[:,:,1]]+fore,2,dim=2)<=1, \
self.one,self.zero)
return fore_result, fore_warp
def forward(self, x, heatmap=None):
"""
x: (batch, c, x_dim, y_dim)
"""
# print(x.shape)
coords = self.coords.numpy().repeat(x.shape[0], 1)
coords = fluid.dygraph.to_variable(coords)
if self.with_boundary and heatmap is not None:
heatmap = heatmap.numpy()
boundary_channel = np.clamp(heatmap[:, -1:, :, :], 0.0, 1.0)
print('boundary_channel', boundary_channel)
zero_tensor = fluid.layers.zeros_like(self.x_coords)
zero_tensor = zero_tensor.numpy()
self.x_coords = self.x_coords.numpy()
self.y_coords = self.y_coords.numpy()
print(self.x_coords)
print(self.y_coords)
# xx_boundary_channel = paddle_where(self.x_coords, heatmap, 0.05)
# yy_boundary_channel = paddle_where(self.y_coords, heatmap, 0.05)
xx_boundary_channel = np.where(boundary_channel > 0.05,
self.x_coords, zero_tensor)
yy_boundary_channel = np.where(boundary_channel > 0.05,
self.y_coords, zero_tensor)
xx_boundary_channel = fluid.dygraph.to_variable(
xx_boundary_channel)
yy_boundary_channel = fluid.dygraph.to_variable(
yy_boundary_channel)
# boundary_channel = fluid.layers.clip(heatmap[:, -1:, :, :], 0.0, 1.0)
# xx_boundary_channel = paddle_where(self.x_coords, boundary_channel, 0.05)
# yy_boundary_channel = paddle_where(self.y_coords, boundary_channel, 0.05)
coords = fluid.layers.concat(
[coords, xx_boundary_channel, yy_boundary_channel], axis=1)
coords = fluid.layers.reshape(
coords, [x.shape[0], 3, coords.shape[2], coords.shape[3]])
x_and_coords = fluid.layers.concat([x, coords], axis=1)
return x_and_coords