def __init__(self, phase, base, extras, head, num_classes):
super(SSD, self).__init__()
self.phase = phase
self.num_classes = num_classes
# TODO: implement __call__ in PriorBox
self.priorbox = PriorBox(v2)
# PyTorch 1.5.1
# self.priors = Variable(self.priorbox.forward(), volatile=True)
self.priors = self.priorbox.forward()
self.size = 300
# SSD network
self.vgg = nn.ModuleList(base)
# Layer learns to scale the l2 normalized features from conv4_3
self.L2Norm = L2Norm(512, 20)
self.extras = nn.ModuleList(extras)
self.loc = nn.ModuleList(head[0])
self.conf = nn.ModuleList(head[1])
if phase == 'test':
# PyTorch 1.5.1
# self.softmax = nn.Softmax()
# self.detect = Detect(num_classes, 0, 200, 0.01, 0.45)
self.softmax = nn.Softmax(dim=-1)
self.detect = Detect()
def __init__(self, phase, base, extras, head, num_classes):
super(SSD, self).__init__()
self.phase = phase
self.num_classes = num_classes
# TODO: implement __call__ in PriorBox
self.priorbox = PriorBox(v2)
self.priors = Variable(self.priorbox.forward(), volatile=True)
self.size = 512
# SSD network
self.vgg = nn.ModuleList(base)
# Layer learns to scale the l2 normalized features from conv4_3
self.L2Norm = L2Norm(512, 20)
self.extras = nn.ModuleList(extras)
# fused conv4_3 and conv5_3
self.conv3_3 = nn.Conv2d(256, 256, 3, 1, 1)
self.conv4_3 = nn.Conv2d(512, 512, 3, 1, 1)
self.deconv = nn.ConvTranspose2d(512, 512, 2, 2)
self.deconv2 = nn.ConvTranspose2d(512, 256, 2, 2)
self.conv5_3 = nn.Conv2d(512, 512, 3, 1, 1)
self.L2Norm5_3 = L2Norm(512, 10)
self.L2Norm3_3 = L2Norm(256, 20)
self.loc = nn.ModuleList(head[0])
self.conf = nn.ModuleList(head[1])
if self.phase == 'test':
self.softmax = nn.Softmax()
self.detect = Detect(num_classes, 0, 300, 0.01, 0.45)
def __init__(self, phase, size, base, extras, head, num_classes):
super(SSD, self).__init__()
self.phase = phase
self.num_classes = num_classes
self.cfg = {
'num_classes': 21,
'lr_steps': (80000, 100000, 120000),
'max_iter': 120000,
'feature_maps': [38, 19, 10, 5, 3, 1],
'min_dim': 300,
'steps': [8, 16, 32, 64, 100, 300],
'min_sizes': [30, 60, 111, 162, 213, 264],
'max_sizes': [60, 111, 162, 213, 264, 315],
'aspect_ratios': [[2], [2, 3], [2, 3], [2, 3], [2], [2]],
'variance': [0.1, 0.2],
'clip': True,
'name': 'VOC',
}
self.priorbox = PriorBox(self.cfg)
with torch.no_grad():
self.priors = Variable(self.priorbox.forward())
self.size = size
# SSD network
self.vgg = nn.ModuleList(base)
self.L2Norm = L2Norm(512, 20)
self.extras = nn.ModuleList(extras)
self.loc = nn.ModuleList(head[0])
self.conf = nn.ModuleList(head[1])
if phase == 'test':
self.softmax = nn.Softmax(dim=-1)
self.detect = Detect(num_classes, 0, 200, 0.01, 0.45)
def __init__(self, phase, base, head, num_classes):
super(MSC, self).__init__()
self.phase = phase
self.num_classes = num_classes
self.priorbox = PriorBox(v2)
self.priors = Variable(self.priorbox.forward(), volatile=True)
self.size = 512
self.vgg = nn.ModuleList(base)
self.L2Norm4_3 = L2Norm(512, 20)
self.L2Norm5_3 = L2Norm(512, 10)
self.deconv7 = nn.ConvTranspose2d(2048, 2048, 2, 2)
self.deconv6 = nn.ConvTranspose2d(2048, 512, 2, 2)
self.deconv5 = nn.ConvTranspose2d(512, 512, 2, 2)
self.conv_fc6 = nn.Conv2d(2048, 2048, 3, 1, 1)
self.conv5_3 = nn.Conv2d(512, 512, 3, 1, 1)
self.conv4_3 = nn.Conv2d(512, 512, 3, 1, 1)
self.loc = nn.ModuleList(head[0])
self.conf = nn.ModuleList(head[1])
if self.phase == 'test':
self.softmax = nn.Softmax()
self.detect = Detect(num_classes, 0, 300, 0.01, 0.45)
def __init__(self, phase, basenet, head, num_classes):
super(TibNet, self).__init__()
self.phase = phase
self.num_classes = num_classes
self.sa = mobilefacenet.SpatialAttention()
self.base = nn.ModuleList(basenet)
self.upfeat = []
for it in range(5):
self.upfeat.append(upsample(in_channels=64, out_channels=64))
self.upfeat = nn.ModuleList(self.upfeat)
self.loc = nn.ModuleList(head[0])
self.conf = nn.ModuleList(head[1])
if self.phase == 'test':
self.softmax = nn.Softmax(dim=-1)
self.detect = Detect(cfg)
def __init__(self, phase, base, head, num_classes):
super(STDN, self).__init__()
self.phase = phase
self.num_classes = num_classes
self.size = 512 # 没有用
self.priorbox = PriorBox(v2)
self.priors = Variable(self.priorbox.forward(), volatile=True)
self.basenet = nn.ModuleList(base)
self.loc = nn.ModuleList(head[0])
self.conf = nn.ModuleList(head[1])
self.out1 = nn.AvgPool2d(9, 9)
self.out2 = nn.AvgPool2d(3, 3)
self.out3 = nn.AvgPool2d(2, 2)
if self.phase == 'test':
self.softmax = nn.Softmax()
self.detect = Detect(num_classes, 0, 300, 0.01, 0.45)
def __init__(self, phase, base, head, num_classes):
super(EXTD, self).__init__()
self.phase = phase
self.num_classes = num_classes
# SSD network
self.base = nn.ModuleList(base)
self.upfeat = []
for it in range(5):
self.upfeat.append(upsample(in_channels=48, out_channels=48))
self.upfeat = nn.ModuleList(self.upfeat)
self.loc = nn.ModuleList(head[0])
self.conf = nn.ModuleList(head[1])
if self.phase == 'test':
self.softmax = nn.Softmax(dim=-1)
self.detect = Detect(cfg)
def __init__(self, phase, base, head, num_classes):
super(EXTD, self).__init__()
self.phase = phase
self.num_classes = num_classes
'''
self.priorbox = PriorBox(size,cfg)
self.priors = Variable(self.priorbox.forward(), volatile=True)
'''
# SSD network
self.base = nn.ModuleList(base)
self.upfeat = []
for it in range(5):
self.upfeat.append(upsample(in_channels=32, out_channels=32))
self.upfeat = nn.ModuleList(self.upfeat)
self.loc = nn.ModuleList(head[0])
self.conf = nn.ModuleList(head[1])
if self.phase == 'test':
self.softmax = nn.Softmax(dim=-1)
self.detect = Detect(cfg)
def __init__(self, phase, size, base, extras, head, num_classes):
super(SSD, self).__init__()
self.phase = phase
self.num_classes = num_classes
self.cfg = voc # (coco, voc)[num_classes == 21]
self.priorbox = PriorBox(self.cfg)
self.priors = self.priorbox.forward(
) # Generate default boxes (anchors)
self.size = size
# SSD network
self.vgg = nn.ModuleList(base)
# Layer learns to scale the l2 normalized features from conv4_3
self.L2Norm = L2Norm(512, 20)
self.extras = nn.ModuleList(extras)
self.loc = nn.ModuleList(head[0])
self.conf = nn.ModuleList(head[1])
if phase == 'test':
self.softmax = nn.Softmax(dim=-1)
self.detect = Detect(num_classes, 0, 200, 0.01, 0.45)
class SSD(nn.Module):
"""Single Shot Multibox Architecture
The network is composed of a base VGG network followed by the
added multibox conv layers. Each multibox layer branches into
1) conv2d for class conf scores
2) conv2d for localization predictions
3) associated priorbox layer to produce default bounding
boxes specific to the layer's feature map size.
See: https://arxiv.org/pdf/1512.02325.pdf for more details.
Args:
phase: (string) Can be "test" or "train"
base: VGG16 layers for input, size of either 300 or 500
extras: extra layers that feed to multibox loc and conf layers
head: "multibox head" consists of loc and conf conv layers
"""
def __init__(self, phase, base, extras, head, num_classes):
super(SSD, self).__init__()
self.phase = phase
self.num_classes = num_classes
# TODO: implement __call__ in PriorBox
self.priorbox = PriorBox(v2)
# PyTorch 1.5.1
# self.priors = Variable(self.priorbox.forward(), volatile=True)
self.priors = self.priorbox.forward()
self.size = 300
# SSD network
self.vgg = nn.ModuleList(base)
# Layer learns to scale the l2 normalized features from conv4_3
self.L2Norm = L2Norm(512, 20)
self.extras = nn.ModuleList(extras)
self.loc = nn.ModuleList(head[0])
self.conf = nn.ModuleList(head[1])
if phase == 'test':
# PyTorch 1.5.1
# self.softmax = nn.Softmax()
# self.detect = Detect(num_classes, 0, 200, 0.01, 0.45)
self.softmax = nn.Softmax(dim=-1)
self.detect = Detect()
def forward(self, x):
"""Applies network layers and ops on input image(s) x.
Args:
x: input image or batch of images. Shape: [batch,3*batch,300,300].
Return:
Depending on phase:
test:
Variable(tensor) of output class label predictions,
confidence score, and corresponding location predictions for
each object detected. Shape: [batch,topk,7]
train:
list of concat outputs from:
1: confidence layers, Shape: [batch*num_priors,num_classes]
2: localization layers, Shape: [batch,num_priors*4]
3: priorbox layers, Shape: [2,num_priors*4]
"""
sources = list()
loc = list()
conf = list()
# apply vgg up to conv4_3 relu
for k in range(23):
x = self.vgg[k](x)
s = self.L2Norm(x)
sources.append(s)
# apply vgg up to fc7
for k in range(23, len(self.vgg)):
x = self.vgg[k](x)
sources.append(x)
# apply extra layers and cache source layer outputs
for k, v in enumerate(self.extras):
x = F.relu(v(x), inplace=True)
if k % 2 == 1:
sources.append(x)
# apply multibox head to source layers
for (x, l, c) in zip(sources, self.loc, self.conf):
loc.append(l(x).permute(0, 2, 3, 1).contiguous())
conf.append(c(x).permute(0, 2, 3, 1).contiguous())
loc = torch.cat([o.view(o.size(0), -1) for o in loc], 1)
conf = torch.cat([o.view(o.size(0), -1) for o in conf], 1)
if self.phase == "test":
# PyTorch 1.5.1
# output = self.detect(
# loc.view(loc.size(0), -1, 4), # loc preds
# self.softmax(conf.view(-1, self.num_classes)), # conf preds
# self.priors.type(type(x.data)) # default boxes
# )
output = self.detect.apply(
self.num_classes,
0,
200,
0.01,
0.45,
loc.view(loc.size(0), -1, 4), # loc preds
self.softmax(conf.view(-1, self.num_classes)), # conf preds
self.priors.type(type(x.data)) # default boxes
)
else:
output = (loc.view(loc.size(0), -1,
4), conf.view(conf.size(0), -1,
self.num_classes), self.priors)
return output
def load_weights(self, base_file):
other, ext = os.path.splitext(base_file)
if ext == '.pkl' or '.pth':
print('Loading weights into state dict...')
self.load_state_dict(
torch.load(base_file,
map_location=lambda storage, loc: storage))
print('Finished!')
else:
print('Sorry only .pth and .pkl files supported.')
def __init__(self, phase, num_classes):
super(BlazeFace, self).__init__()
self.firstconv = nn.Sequential(
nn.Conv2d(in_channels=2,
out_channels=16,
kernel_size=3,
stride=2,
padding=1),
nn.BatchNorm2d(16),
nn.ReLU(inplace=True),
)
self.blazeBlock = nn.Sequential(
BlazeBlock(in_channels=16, out_channels=16),
# BlazeBlock(in_channels=24, out_channels=24),
BlazeBlock(in_channels=16, out_channels=32, stride=2),
# BlazeBlock(in_channels=48, out_channels=48),
BlazeBlock(in_channels=32, out_channels=32),
)
self.doubleBlazeBlock = nn.Sequential(
DoubleBlazeBlock(in_channels=32,
out_channels=64,
mid_channels=16,
stride=2),
# DoubleBlazeBlock(in_channels=96, out_channels=96, mid_channels=24),
DoubleBlazeBlock(in_channels=64, out_channels=64, mid_channels=16),
)
self.conv2d_8x8_classificators = nn.Sequential(
nn.Conv2d(in_channels=64, out_channels=6, kernel_size=1, stride=1),
nn.BatchNorm2d(6),
)
self.conv2d_16x16_classificators = nn.Sequential(
nn.Conv2d(in_channels=64, out_channels=2, kernel_size=1, stride=1),
nn.BatchNorm2d(2),
)
self.conv2d_8x8_regressors = nn.Sequential(
nn.Conv2d(in_channels=64, out_channels=64, kernel_size=1,
stride=1),
nn.BatchNorm2d(64),
)
self.conv2d_16x16_regressors = nn.Sequential(
nn.Conv2d(in_channels=64, out_channels=32, kernel_size=1,
stride=1),
nn.BatchNorm2d(32),
)
self.phase = phase
self.cfg = celeba
self.priorbox = PriorBox(self.cfg)
self.num_classes = num_classes
# self.priors = Variable(self.priorbox.forward(), volatile=True)
if phase == 'test':
self.softmax = nn.Softmax(dim=-1)
self.detect = Detect(self.num_classes, 0, 200, 0.01, 0.45)
with torch.no_grad():
self.priors = self.priorbox.forward()
print(self.priors.shape)
self.initialize()
def main(input_path, output_path, detection_model_path='weights/WIDERFace_DSFD_RES152.pt'):
cuda = True
torch.set_grad_enabled(False)
device = torch.device('cuda:{}'.format(0))
if cuda and torch.cuda.is_available():
torch.set_default_tensor_type('torch.cuda.FloatTensor')
else:
torch.set_default_tensor_type('torch.FloatTensor')
# Initialize detection model
# cfg = widerface_640
# thresh = cfg['conf_thresh']
# net = build_ssd('test', cfg['min_dim'], cfg['num_classes']) # initialize SSD
# net.load_state_dict(torch.load(detection_model_path))
# net = net.cuda()
# net.eval()
cfg = widerface_640
thresh = cfg['conf_thresh']
net = torch.jit.load(detection_model_path, map_location=device)
net.eval()
print('Finished loading detection model!')
transform = TestBaseTransform((104, 117, 123))
detect = Detect(cfg['num_classes'], 0, cfg['num_thresh'], cfg['conf_thresh'], cfg['nms_thresh'])
# Open target video file
cap = cv2.VideoCapture(input_path)
if not cap.isOpened():
raise RuntimeError('Failed to read video: ' + input_path)
total_frames = int(cap.get(cv2.CAP_PROP_FRAME_COUNT))
fps = cap.get(cv2.CAP_PROP_FPS)
target_vid_width = int(cap.get(cv2.CAP_PROP_FRAME_WIDTH))
target_vid_height = int(cap.get(cv2.CAP_PROP_FRAME_HEIGHT))
# Calculate priors
image_size = (target_vid_height, target_vid_width)
featuremap_size = [(math.ceil(image_size[0] / (2 ** (2 + i))), math.ceil(image_size[1] / (2 ** (2 + i))))
for i in range(6)]
priors = get_prior_boxes(cfg, featuremap_size, image_size).to(device)
# Initialize output video file
if output_path is not None:
if os.path.isdir(output_path):
output_filename = os.path.splitext(os.path.basename(input_path))[0] + '.mp4'
output_path = os.path.join(output_path, output_filename)
fourcc = cv2.VideoWriter_fourcc(*'x264')
out_vid = cv2.VideoWriter(output_path, fourcc, fps, (target_vid_width, target_vid_height))
else:
out_vid = None
#
max_im_shrink = ((2000.0 * 2000.0) / (target_vid_height * target_vid_width)) ** 0.5
shrink = max_im_shrink if max_im_shrink < 1 else 1
# For each frame in the video
for i in tqdm(range(total_frames)):
ret, frame = cap.read()
if frame is None:
continue
# Process
frame_tensor = torch.from_numpy(transform(frame)[0]).permute(2, 0, 1).unsqueeze(0).to(device)
# image_size = frame_tensor.shape[2:]
# featuremap_size = [(math.ceil(image_size[0] / (2 ** (2 + i))), math.ceil(image_size[1] / (2 ** (2 + i))))
# for i in range(6)]
# priors = get_prior_boxes(cfg, featuremap_size, image_size).to(device)
pred = net(frame_tensor)
detections = detect(pred[:, :, :4], pred[:, :, 4:], priors)
det = []
shrink = 1.0
scale = torch.Tensor([image_size[1] / shrink, image_size[0] / shrink,
image_size[1] / shrink, image_size[0] / shrink])
for i in range(detections.size(1)):
j = 0
while detections[0, i, j, 0] >= thresh:
curr_det = detections[0, i, j, [1, 2, 3, 4, 0]].cpu().numpy()
curr_det[:4] *= scale.cpu().numpy()
det.append(curr_det)
j += 1
# curr_det[:4] *= scale
# score = detections[0, i, j, 0]
# # label_name = labelmap[i-1]
# pt = (detections[0, i, j, 1:] * scale).cpu().numpy()
# coords = (pt[0], pt[1], pt[2], pt[3])
# det.append([pt[0], pt[1], pt[2], pt[3], score])
# j += 1
det = np.row_stack((det))
# if det.shape[0] > 1:
# det = bbox_vote(det.astype(float))
det = np.round(det[det[:, 4] > 0.5, :4]).astype(int)
# # Render
render_img = frame
for rect in det:
# cv2.rectangle(render_img, tuple(rect[:2]), tuple(rect[:2] + rect[2:]), (0, 0, 255), 1)
cv2.rectangle(render_img, tuple(rect[:2]), tuple(rect[2:]), (0, 0, 255), 1)
if out_vid is not None:
out_vid.write(render_img)
cv2.imshow('render_img', render_img)
if cv2.waitKey(1) & 0xFF == ord('q'):
break