def expand_resize_data(prediction=None,
submission_size=(3384, 1710),
offset=690):
pred_mask = decode_labels(prediction)
expand_mask = cv2.resize(pred_mask,
(submission_size[0], submission_size[1] - offset),
interpolation=cv2.INTER_NEAREST)
submission_mask = np.zeros((submission_size[1], submission_size[0]),
dtype='uint8')
submission_mask[offset:, :] = expand_mask
return submission_mask
TA_sum=[]
for i,j in result["TP"].items():
TP_sum.append(j)
for i,j in result["TA"].items():
TA_sum.append(j)
TP_sum=np.array(TP_sum)
TA_sum=np.array(TA_sum)
TP_sum=TP_sum[1:].sum()
TA_sum=TA_sum[1:].sum()
acc='%.5f' %(TP_sum/TA_sum)
print("acc:",acc)
acc_all.append(acc)
pred = pred.detach().cpu().numpy()
mask = np.squeeze(mask.detach().cpu().numpy())
for k in range(batch_size):
pred_=pred[k,:]
pred_ = decode_labels(np.squeeze(pred_))
mask_=mask[k,:]
mp.imsave('./predict/pred/' + str(j*batch_size+k) + '.png', pred_)
mp.imsave('./predict/label/' + str(j*batch_size+k) + '.png', mask_)
j=j+1
print("test_miou_Avr:", np.mean(test_all_miou))
acc_all = np.array(acc_all, dtype=np.float32)
print("acc_Avr:", np.mean(acc_all))
TA_sum.append(j)
TP_sum = np.array(TP_sum)
TA_sum = np.array(TA_sum)
TP_sum = TP_sum[1:].sum()
TA_sum = TA_sum[1:].sum()
acc = '%.5f' % (TP_sum / TA_sum)
print("acc:", acc)
acc_all.append(acc)
pred = pred.detach().cpu().numpy()
mask = mask.detach().cpu().numpy()
# print("pred.shape:",pred.shape)
# print("mask.shape:",mask.shape)
for k in range(batch_size):
pred_ = pred[k, :]
pred_ = decode_labels(pred_)
mask_ = mask[k, :]
# print("pred_.shape:",pred_.shape)
# print("mask_.shape:",mask_.shape)
mp.imsave('./predict/pred/' + str(j * batch_size + k) + '.png',
pred_)
mp.imsave(
'./predict/label/' + str(j * batch_size + k) + '.png',
mask_)
j = j + 1
print("test_miou_Avr:", np.mean(test_all_miou))
acc_all = np.array(acc_all, dtype=np.float32)
print("acc_Avr:", np.mean(acc_all))