def prepare_testing_init(box, score):
if cfg.ex_type == 'corner':
i_it_4pys = snake_decode.get_init(box, init_type='corner')
else:
i_it_4pys = snake_decode.get_init(box)
init_boxes = i_it_4pys
i_it_4pys = uniform_upsample(i_it_4pys, snake_config.init_poly_num)
c_it_4pys = img_poly_to_can_poly(i_it_4pys)
if 0:
print('--------- vis init-pts ---------')
import matplotlib.pyplot as plt
boxes = box.cpu().numpy()[0, ...]
init_boxes = init_boxes.cpu().numpy()[0, ...]
init_polys = i_it_4pys.cpu().numpy()[0, ...]
I = np.zeros((100, 100))
plt.imshow(I)
for k in range(1):
box = boxes[k]
init_poly = init_polys[k, ...]
init_box = init_boxes[k, ...]
init_poly = np.concatenate((init_poly, init_poly[0:1, :]), axis=0)
init_box = np.concatenate((init_box, init_box[0:1, :]), axis=0)
plt.plot(init_box[:, 0], init_box[:, 1])
plt.plot(init_poly[:15, 0], init_poly[:15, 1], 'r+')
plt.show()
exit()
if 0:
print('score.shape:', score.shape)
print('i_it_4py.shape:', i_it_4pys.shape)
ind = score > snake_config.ct_score
if 0:
print('ind:', ind)
print('ind.shape:', ind.size())
for k in range(ind.size(0)):
a = torch.full([ind[k].sum()], k)
print('a:', a)
i_it_4pys = i_it_4pys[ind]
c_it_4pys = c_it_4pys[ind]
ind = torch.cat(
[torch.full([ind[i].sum()], i) for i in range(ind.size(0))], dim=0)
if 0:
print('ind:', ind)
exit()
init = {'i_it_4py': i_it_4pys, 'c_it_4py': c_it_4pys, 'ind': ind}
return init
def prepare_training_box(ret, batch, init):
box = ret['detection'][..., :4]
score = ret['detection'][..., 4]
batch_size = box.size(0)
i_gt_4py = batch['i_gt_4py']
ct_01 = batch['ct_01'].byte()
ind = [
get_box_match_ind(box[i], score[i], i_gt_4py[i][ct_01[i]])
for i in range(batch_size)
]
box_ind = [ind_[0] for ind_ in ind]
gt_ind = [ind_[1] for ind_ in ind]
i_it_4py = torch.cat([
snake_decode.get_init(box[i][box_ind[i]][None])
for i in range(batch_size)
],
dim=1)
if i_it_4py.size(1) == 0:
return
i_it_4py = uniform_upsample(i_it_4py, snake_config.init_poly_num)[0]
c_it_4py = img_poly_to_can_poly(i_it_4py)
i_gt_4py = torch.cat(
[batch['i_gt_4py'][i][gt_ind[i]] for i in range(batch_size)], dim=0)
c_gt_4py = torch.cat(
[batch['c_gt_4py'][i][gt_ind[i]] for i in range(batch_size)], dim=0)
init_4py = {
'i_it_4py': i_it_4py,
'c_it_4py': c_it_4py,
'i_gt_4py': i_gt_4py,
'c_gt_4py': c_gt_4py
}
i_it_py = snake_decode.get_octagon(i_gt_4py[None])
i_it_py = uniform_upsample(i_it_py, snake_config.poly_num)[0]
c_it_py = img_poly_to_can_poly(i_it_py)
i_gt_py = torch.cat(
[batch['i_gt_py'][i][gt_ind[i]] for i in range(batch_size)], dim=0)
init_py = {'i_it_py': i_it_py, 'c_it_py': c_it_py, 'i_gt_py': i_gt_py}
ind = torch.cat(
[torch.full([len(gt_ind[i])], i) for i in range(batch_size)], dim=0)
if snake_config.train_pred_box_only:
for k, v in init_4py.items():
init[k] = v
for k, v in init_py.items():
init[k] = v
init['4py_ind'] = ind
init['py_ind'] = ind
else:
init.update(
{k: torch.cat([init[k], v], dim=0)
for k, v in init_4py.items()})
init.update({'4py_ind': torch.cat([init['4py_ind'], ind], dim=0)})
init.update(
{k: torch.cat([init[k], v], dim=0)
for k, v in init_py.items()})
init.update({'py_ind': torch.cat([init['py_ind'], ind], dim=0)})
def prepare_testing_init(box, score):
i_it_4pys = snake_decode.get_init(box)
i_it_4pys = uniform_upsample(i_it_4pys, snake_config.init_poly_num)
c_it_4pys = img_poly_to_can_poly(i_it_4pys)
ind = score > snake_config.ct_score
i_it_4pys = i_it_4pys[ind]
c_it_4pys = c_it_4pys[ind]
ind = torch.cat([torch.full([ind[i].sum()], i) for i in range(ind.size(0))], dim=0)
init = {'i_it_4py': i_it_4pys, 'c_it_4py': c_it_4pys, 'ind': ind}
return init
def prepare_testing_init(self, output):
i_it_4py = snake_decode.get_init(output['cp_box'][None])
i_it_4py = snake_gcn_utils.uniform_upsample(i_it_4py,
snake_config.init_poly_num)
c_it_4py = snake_gcn_utils.img_poly_to_can_poly(i_it_4py)
i_it_4py = i_it_4py[0]
c_it_4py = c_it_4py[0]
ind = output['roi_ind'][output['cp_ind'].long()]
init = {'i_it_4py': i_it_4py, 'c_it_4py': c_it_4py, 'ind': ind}
output.update({'it_ex': init['i_it_4py']})
return init
def prepare_training_box(ret, batch, init): #这里准备的是用于训练的bbox框
box = ret['detection'][..., :4]
# tmp_file = open('/home/tianhao.lu/code/Deep_snake/snake/Result/Contour/contour.log', 'w')
# tmp_file.writelines("bbox:" + str(box) + "\n")
# tmp_file.writelines(str("*************************************************************** \n"))
# tmp_file.close()
score = ret['detection'][..., 4]
batch_size = box.size(0)
i_gt_4py = batch['i_gt_4py']
ct_01 = batch['ct_01'].byte()
ind = [get_box_match_ind(box[i], score[i], i_gt_4py[i][ct_01[i]]) for i in range(batch_size)]
box_ind = [ind_[0] for ind_ in ind]
gt_ind = [ind_[1] for ind_ in ind]
i_it_4py = torch.cat([snake_decode.get_init(box[i][box_ind[i]][None]) for i in range(batch_size)], dim=1)
if i_it_4py.size(1) == 0:
return
i_it_4py = uniform_upsample(i_it_4py, snake_config.init_poly_num)[0]
c_it_4py = img_poly_to_can_poly(i_it_4py)
i_gt_4py = torch.cat([batch['i_gt_4py'][i][gt_ind[i]] for i in range(batch_size)], dim=0)
c_gt_4py = torch.cat([batch['c_gt_4py'][i][gt_ind[i]] for i in range(batch_size)], dim=0)
init_4py = {'i_it_4py': i_it_4py, 'c_it_4py': c_it_4py, 'i_gt_4py': i_gt_4py, 'c_gt_4py': c_gt_4py}
i_it_py = snake_decode.get_octagon(i_gt_4py[None])
i_it_py = uniform_upsample(i_it_py, snake_config.poly_num)[0]
c_it_py = img_poly_to_can_poly(i_it_py)
i_gt_py = torch.cat([batch['i_gt_py'][i][gt_ind[i]] for i in range(batch_size)], dim=0)
init_py = {'i_it_py': i_it_py, 'c_it_py': c_it_py, 'i_gt_py': i_gt_py}
ind = torch.cat([torch.full([len(gt_ind[i])], i) for i in range(batch_size)], dim=0)
if snake_config.train_pred_box_only:
for k, v in init_4py.items():
init[k] = v
for k, v in init_py.items():
init[k] = v
init['4py_ind'] = ind
init['py_ind'] = ind
else:
init.update({k: torch.cat([init[k], v], dim=0) for k, v in init_4py.items()})
init.update({'4py_ind': torch.cat([init['4py_ind'], ind], dim=0)})
init.update({k: torch.cat([init[k], v], dim=0) for k, v in init_py.items()})
init.update({'py_ind': torch.cat([init['py_ind'], ind], dim=0)})
def prepare_testing_init(box, score):
i_it_4pys = snake_decode.get_init(box)
i_it_4pys = uniform_upsample(i_it_4pys, snake_config.init_poly_num)
c_it_4pys = img_poly_to_can_poly(i_it_4pys)
#ind=[batch][t,f,t,f...]
ind = score > snake_config.ct_score
i_it_4pys = i_it_4pys[ind]
c_it_4pys = c_it_4pys[ind]
#torch.full([ind[i].sum()], i)->[i,i...i] 长度为这张图片score > snake_config.ct_score的个数
# ind= [0,0,0...,0]
# [1,1,1...,1]
# [2,2,2...,2]
# [3,3,3...,3]
#ind[n]为第n张图片的...
ind = torch.cat(
[torch.full([ind[i].sum()], i) for i in range(ind.size(0))], dim=0)
init = {'i_it_4py': i_it_4pys, 'c_it_4py': c_it_4pys, 'ind': ind}
return init