-- A Pytorch implementing of A Deep Learning approach to Template Matching
Require one template image and one source image.
Template image needs to be padded to the same size as source image.
And then resize to (512, 512) ...
from model import TemplateMatching
t = torch.randn(4, 3, 512, 512).cuda()
x = torch.randn(4, 3, 512, 512).cuda()
net = TemplateMatching(
z_dim=64, output_channel=512, pretrain=False).cuda()
# checkpoint
checkpoint = torch.load('path/to/checkpoint/model_best.pth) # output/checkpoint/
net.load_state_dict(checkpoint['TemplateMatching'])
res = net(x, t) # binary map Require one source image and a folder containing all of the probable apperences of target.
Return the location of target (only the most likely one).
from ccoeff import template_matching
template_dir = 'data/templates/pacman/'
img = 'source.png'
res = template_matching(img, template_dir,
vis=False, return_ori=False)