def calc_basic_test():
band_7_test = utils.get_truth(get_swir2(), 0.03, '>')
btc_test = utils.get_truth(utils.convert_to_celsius(get_temp()), 27.0, '<')
ndsi_test = utils.get_truth(calc_ndsi(), 0.8, '<')
ndvi_test = utils.get_truth(calc_ndvi(), 0.8, '<')
basic_test = np.logical_and.reduce((band_7_test,
btc_test,
ndsi_test,
ndvi_test))
return basic_test
print 'sideoutput:', img_path, MSE_loss
MSE_loss_list.append(MSE_loss)
if save_test_img:
pred_img = cv2.resize(new_Image.astype(np.uint8), (ROI_img.shape[1], ROI_img.shape[0]))
tmp_img = pred_img.copy()
tmp_ROI_img = ROI_img.copy()
pred_img[pred_img==3]=2
pred_img[NucluesImage==1]=3
ROI_img = crop_boundry(ROI_img, pred_img)
alpha_img = ori_img.copy()
alpha_img[BeginY:, ally[0]:ally[1]] = ROI_img
ROI_img = cv2.addWeighted(alpha_img, 0.4, ori_img, 0.6, 0)
## imshow the truth
ROI_img = get_truth(ROI_img, img_path, ally, BeginY)
if not os.path.exists(os.path.join(args.results,model_name)):
os.mkdir(os.path.join(args.results,model_name))
img_path2 = img_path[:-4] + '_ray.png'
save_name = os.path.join(args.results,model_name, img_path2)
cv2.imwrite(save_name, ROI_img)
FullImage = np.zeros(imshape)
FinalShape = FinalShape * 1
pred_imgFinal = pred_img
FullImage[BeginY:, ally[0]:ally[1]] = pred_imgFinal
a = np.zeros(imshape)
tmp_label = FullImage.copy()
# with open('./logs/data/%s_ray.pkl'%img_path, 'w+') as f:
# data = [a, tmp_label, img_path, ally]
# pkl.dump(data, f)
new_Image = ppi.astype(np.uint8)
kernel = cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (3, 3))
new_Image = cv2.morphologyEx(new_Image, cv2.MORPH_CLOSE, kernel)
ori_img_path = os.path.join(ori_data, path)
ori_img = cv2.imread(ori_img_path)
ROI_img = ori_img[top:, left:right, :].copy()
pred_img = cv2.resize(new_Image.astype(np.uint8),
(ROI_img.shape[1], ROI_img.shape[0]),
interpolation=cv2.INTER_LANCZOS4)
ROI_img = crop_boundry(ROI_img, pred_img)
alpha_img = ori_img.copy()
alpha_img[top:, left:right] = ROI_img
ROI_img = cv2.addWeighted(alpha_img, 0.4, ori_img, 0.6, 0)
try:
ROI_img = get_truth(ROI_img, path, [left, right], top)
except:
print path
if not os.path.exists('./visual_results/%s' % model_name):
os.mkdir('./visual_results/%s' % model_name)
save_name = './visual_results/%s/label_%s_%s.png' % (model_name,
epoch, i)
cv2.imwrite(save_name, ROI_img)
MSE_loss = None
if not os.path.exists('./models/%s' % model_name):
os.mkdir('./models/%s' % model_name)
if args.test_every_step:
start = time.time()
print("Begining test ")
def calc_whiteness_test():
whiteness_test = utils.get_truth(calc_whiteness(), 0.7, '<') # 0.7 in paper
return whiteness_test
del pos_list, pos_lab, neg_list, neg_lab
# hog_par = {'winStride': (8, 8), 'padding': (0, 0), 'scale': 1.2}
# hog_par = {'hitThreshold': 1.2, 'winStride': (8, 8), 'padding': (0, 0), 'scale': 1.2, 'finalThreshold': 4}
hog_par = {'hitThreshold': 1.4, 'winStride': (8, 8), 'padding': (0, 0), 'scale': 1.2, 'finalThreshold': 2}
"""Initialize and train the SVM"""
x, y = utils.get_data(hog_list)
del hog_list
svm_par = dict(kernel_type=cv2.SVM_LINEAR, svm_type=cv2.SVM_C_SVC)
# svm_par = dict(kernel_type=cv2.SVM_LINEAR, svm_type=cv2.SVM_C_SVC, C=0.01)
svm_vec = utils.load_svm("svm.pickle", x, y, svm_par)
# svm_vec = cv2.HOGDescriptor_getDefaultPeopleDetector()
hog_obj.setSVMDetector(np.array(svm_vec))
"""Ground truth"""
true_detection_list = utils.get_truth('view8.json')
n_frame = len(true_detection_list)
"""Multi-scale detector on the video"""
hog_detection_list = []
frm_list = utils.get_frame(frm_path)
frm_list = frm_list[0:n_frame]
i = -1
for frm in frm_list:
i += 1
found_true = true_detection_list[i]
found_filtered = []
found, w = hog_obj.detectMultiScale(frm, **hog_par)
for r in found:
inside = False
for q in found:
new_Image = ppi.astype(np.uint8)
kernel = cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (3, 3))
new_Image = cv2.morphologyEx(new_Image, cv2.MORPH_CLOSE, kernel)
ori_img_path = os.path.join(ori_data, path)
ori_img = cv2.imread(ori_img_path)
ROI_img = ori_img[top:, left:right, :].copy()
pred_img = cv2.resize(new_Image.astype(np.uint8),
(ROI_img.shape[1], ROI_img.shape[0]),
interpolation=cv2.INTER_LANCZOS4)
ROI_img = crop_boundry(ROI_img, pred_img)
alpha_img = ori_img.copy()
alpha_img[top:, left:right] = ROI_img
ROI_img = cv2.addWeighted(alpha_img, 0.4, ori_img, 0.6, 0)
ROI_img = get_truth(ROI_img, path, [left, right], top)
if not os.path.exists('./visual_results/%s' % model_name):
os.mkdir('./visual_results/%s' % model_name)
save_name = './visual_results/%s/label_%s_%s.png' % (model_name,
epoch, i)
cv2.imwrite(save_name, ROI_img)
MSE_loss = None
if not os.path.exists('./models/%s' % model_name):
os.mkdir('./models/%s' % model_name)
if epoch and epoch % 10 == 0:
torch.save(model.state_dict(),
'./models/%s/%s_%s.pth' % (model_name, epoch, args.lr))
print 'success save every step model'
'hitThreshold': 1.4,
'winStride': (8, 8),
'padding': (0, 0),
'scale': 1.2,
'finalThreshold': 2
}
"""Initialize and train the SVM"""
x, y = utils.get_data(hog_list)
del hog_list
svm_par = dict(kernel_type=cv2.SVM_LINEAR, svm_type=cv2.SVM_C_SVC)
# svm_par = dict(kernel_type=cv2.SVM_LINEAR, svm_type=cv2.SVM_C_SVC, C=0.01)
svm_vec = utils.load_svm("svm.pickle", x, y, svm_par)
# svm_vec = cv2.HOGDescriptor_getDefaultPeopleDetector()
hog_obj.setSVMDetector(np.array(svm_vec))
"""Ground truth"""
true_detection_list = utils.get_truth('view8.json')
n_frame = len(true_detection_list)
"""Multi-scale detector on the video"""
hog_detection_list = []
frm_list = utils.get_frame(frm_path)
frm_list = frm_list[0:n_frame]
i = -1
for frm in frm_list:
i += 1
found_true = true_detection_list[i]
found_filtered = []
found, w = hog_obj.detectMultiScale(frm, **hog_par)
for r in found:
inside = False
for q in found:
if utils.is_inside(r, q):
