def main():
args = parser.parse_args()
time_stamp = "{0:%Y%m%d-%H%M%S}".format(datetime.now())
save_name = os.path.join(args.save_dir, "train_{}".format(time_stamp))
if not(os.path.isdir(args.save_dir)):
os.makedirs(args.save_dir)
if args.path == None:
raise OSError("path to annotation file must be required.")
C = config.Config()
C.config_filename = save_name + "_config.pickle"
C.model_path = save_name + "_model.hdf5"
C.use_horizontal_flips = bool(args.horizontal_flips)
C.use_vertical_flips = bool(args.vertical_flips)
C.rot_90 = bool(args.rot_90)
all_imgs, classes_count, class_mapping = get_data(args.path)
C.class_mapping = class_mapping
with open(C.config_filename, 'wb') as config_f:
pickle.dump(C,config_f)
print("-------------------------------")
print('path to config file : {}'.format(C.config_filename))
print("-------------------------------")
train_imgs = [s for s in all_imgs if s['imageset'] == 'trainval']
val_imgs = [s for s in all_imgs if s['imageset'] == 'test']
data_gen_train = data_generators.get_anchor_gt(train_imgs, classes_count, C, K.image_dim_ordering(), mode='train')
data_gen_val = data_generators.get_anchor_gt(val_imgs, classes_count, C, K.image_dim_ordering(), mode='val')
model_rpn, model_classifier, model_all = faster_rcnn.get_model(C, classes_count)
losses = np.zeros((args.n_iters, 5))
rpn_accuracy_rpn_monitor, rpn_accuracy_for_epoch = [], []
best_loss = np.Inf
with open('out.csv', 'w') as f:
f.write('Accuracy,RPN classifier,RPN regression,Detector classifier,Detector regression,Total')
f.write('\t')
iter_num = 0
t0 = start_time = time.time()
try:
for epoch_num in range(args.n_epochs):
progbar = generic_utils.Progbar(args.n_iters)
print('Epoch {}/{}'.format(epoch_num + 1, args.n_epochs))
while True:
try:
if len(rpn_accuracy_rpn_monitor) == args.n_iters and C.verbose:
mean_overlapping_bboxes = float(sum(rpn_accuracy_rpn_monitor))/len(rpn_accuracy_rpn_monitor)
rpn_accuracy_rpn_monitor = []
print('Average number of overlapping bounding boxes from RPN = {} for {} previous iterations'.format(mean_overlapping_bboxes, args.n_iters))
if mean_overlapping_bboxes == 0:
print('RPN is not producing bounding boxes that overlap the ground truth boxes. Check RPN settings or keep training.')
X, Y, img_data = next(data_gen_train)
loss_rpn = model_rpn.train_on_batch(X, Y)
P_rpn = model_rpn.predict_on_batch(X)
R = roi_helpers.rpn_to_roi(P_rpn[0], P_rpn[1], C, K.image_dim_ordering(), use_regr=True, overlap_thresh=0.7, max_boxes=300)
# note: calc_iou converts from (x1,y1,x2,y2) to (x,y,w,h) format
X2, Y1, Y2 = roi_helpers.calc_iou(R, img_data, C, class_mapping)
neg_samples = np.where(Y1[0, :, -1] == 1)
pos_samples = np.where(Y1[0, :, -1] == 0)
if len(neg_samples) > 0:
neg_samples = neg_samples[0]
else:
neg_samples = []
if len(pos_samples) > 0:
pos_samples = pos_samples[0]
else:
pos_samples = []
rpn_accuracy_rpn_monitor.append(len(pos_samples))
rpn_accuracy_for_epoch.append((len(pos_samples)))
if len(pos_samples) < C.num_rois//2:
selected_pos_samples = pos_samples.tolist()
else:
selected_pos_samples = np.random.choice(pos_samples, C.num_rois//2, replace=False).tolist()
try:
selected_neg_samples = np.random.choice(neg_samples, C.num_rois - len(selected_pos_samples), replace=False).tolist()
except:
selected_neg_samples = np.random.choice(neg_samples, C.num_rois - len(selected_pos_samples), replace=True).tolist()
sel_samples = selected_pos_samples + selected_neg_samples
loss_class = model_classifier.train_on_batch([X, X2[:, sel_samples, :]], [Y1[:, sel_samples, :], Y2[:, sel_samples, :]])
if iter_num == args.n_iters:
loss_rpn_cls = np.mean(losses[:, 0])
loss_rpn_regr = np.mean(losses[:, 1])
loss_class_cls = np.mean(losses[:, 2])
loss_class_regr = np.mean(losses[:, 3])
class_acc = np.mean(losses[:, 4])
mean_overlapping_bboxes = float(sum(rpn_accuracy_for_epoch)) / len(rpn_accuracy_for_epoch)
rpn_accuracy_for_epoch = []
if C.verbose:
print('Mean number of bounding boxes from RPN overlapping ground truth boxes: {}'.format(mean_overlapping_bboxes))
print('Classifier accuracy for bounding boxes from RPN: {}'.format(class_acc))
print('Loss RPN classifier: {}'.format(loss_rpn_cls))
print('Loss RPN regression: {}'.format(loss_rpn_regr))
print('Loss Detector classifier: {}'.format(loss_class_cls))
print('Loss Detector regression: {}'.format(loss_class_regr))
print('Elapsed time: {}[s]'.format(time.time() - start_time))
target_text_file = open('out.csv', 'a')
target_text_file.write('{},{},{},{},{},{}'.format(class_acc, loss_rpn_cls,
loss_rpn_regr, loss_class_cls, loss_class_regr,
loss_rpn_cls + loss_rpn_regr + loss_class_cls + loss_class_regr))
target_text_file.write('\t')
curr_loss = loss_rpn_cls + loss_rpn_regr + loss_class_cls + loss_class_regr
iter_num = 0
start_time = time.time()
if curr_loss < best_loss:
if C.verbose:
print('Total loss decreased from {} to {}, saving weights'.format(best_loss,curr_loss))
best_loss = curr_loss
model_all.save_weights(C.model_path)
break
losses[iter_num, 0] = loss_rpn[1]
losses[iter_num, 1] = loss_rpn[2]
losses[iter_num, 2] = loss_class[1]
losses[iter_num, 3] = loss_class[2]
losses[iter_num, 4] = loss_class[3]
iter_num += 1
progbar.update(iter_num, [('rpn_cls', np.mean(losses[:iter_num, 0])), ('rpn_regr', np.mean(losses[:iter_num, 1])),
('detector_cls', np.mean(losses[:iter_num, 2])), ('detector_regr', np.mean(losses[:iter_num, 3]))])
except Exception as e:
print('Exception: {}'.format(e))
continue
except KeyboardInterrupt:
t1 = time.time()
print('\nIt took {:.2f}s'.format(t1-t0))
sys.exit('Keyboard Interrupt')
print("training is done")
print("-------------------------------")
print('path to config file : {}'.format(C.config_filename))
print("-------------------------------")
def detect_predict(pic,
C,
model_rpn,
model_classifier,
model_classifier_only,
class_mapping,
class_to_color,
print_dets=False,
export=False):
"""
Detect and predict object in the picture
:param pic: picture numpy array
:param C: config object
:params model_*: models from get_models function
:params class_*: mapping and colors, need to be loaded to keep the same colors/classes
:return: picture with bounding boxes
"""
img = pic
X, ratio = format_img(img, C)
img_scaled = np.transpose(X.copy()[0, (2, 1, 0), :, :], (1, 2, 0)).copy()
img_scaled[:, :, 0] += 123.68
img_scaled[:, :, 1] += 116.779
img_scaled[:, :, 2] += 103.939
img_scaled = img_scaled.astype(np.uint8)
if K.image_data_format() == 'channels_last':
X = np.transpose(X, (0, 2, 3, 1))
# get the feature maps and output from the RPN
[Y1, Y2, F] = model_rpn.predict(X)
R = roi_helpers.rpn_to_roi(Y1,
Y2,
C,
K.image_data_format(),
overlap_thresh=0.7)
# convert from (x1,y1,x2,y2) to (x,y,w,h)
R[:, 2] -= R[:, 0]
R[:, 3] -= R[:, 1]
# apply the spatial pyramid pooling to the proposed regions
bboxes = {}
probs = {}
# print(class_mapping)
for jk in range(R.shape[0] // C.num_rois + 1):
ROIs = np.expand_dims(R[C.num_rois * jk:C.num_rois * (jk + 1), :],
axis=0)
if ROIs.shape[1] == 0:
break
if jk == R.shape[0] // C.num_rois:
#pad R
curr_shape = ROIs.shape
target_shape = (curr_shape[0], C.num_rois, curr_shape[2])
ROIs_padded = np.zeros(target_shape).astype(ROIs.dtype)
ROIs_padded[:, :curr_shape[1], :] = ROIs
ROIs_padded[0, curr_shape[1]:, :] = ROIs[0, 0, :]
ROIs = ROIs_padded
[P_cls, P_regr] = model_classifier_only.predict([F, ROIs])
for ii in range(P_cls.shape[1]):
if np.max(P_cls[0, ii, :]) < bbox_threshold or np.argmax(
P_cls[0, ii, :]) == (P_cls.shape[2] - 1):
continue
cls_name = class_mapping[np.argmax(P_cls[0, ii, :])]
if cls_name not in bboxes:
bboxes[cls_name] = []
probs[cls_name] = []
(x, y, w, h) = ROIs[0, ii, :]
cls_num = np.argmax(P_cls[0, ii, :])
try:
(tx, ty, tw, th) = P_regr[0, ii, 4 * cls_num:4 * (cls_num + 1)]
tx /= C.classifier_regr_std[0]
ty /= C.classifier_regr_std[1]
tw /= C.classifier_regr_std[2]
th /= C.classifier_regr_std[3]
x, y, w, h = roi_helpers.apply_regr(x, y, w, h, tx, ty, tw, th)
except:
pass
bboxes[cls_name].append([
C.rpn_stride * x, C.rpn_stride * y, C.rpn_stride * (x + w),
C.rpn_stride * (y + h)
])
probs[cls_name].append(np.max(P_cls[0, ii, :]))
all_dets = []
boxes_export = {}
for key in bboxes:
bbox = np.array(bboxes[key])
# Eliminating redundant object detection windows
new_boxes, new_probs = roi_helpers.non_max_suppression_fast(
bbox, np.array(probs[key]), overlap_thresh=overlap_thresh)
# Keep only the best prediction per character
jk = np.argmax(new_probs)
# Threshold for best prediction
if new_probs[jk] > 0.55:
(x1, y1, x2, y2) = new_boxes[jk, :]
# Convert predicted picture box coordinates to real-size picture coordinates
(real_x1, real_y1, real_x2,
real_y2) = get_real_coordinates(ratio, x1, y1, x2, y2)
# Exporting box coordinates instead of draw on the picture
if export:
boxes_export[key] = [(real_x1, real_y1, real_x2, real_y2),
int(100 * new_probs[jk])]
else:
cv2.rectangle(
img, (real_x1, real_y1), (real_x2, real_y2),
(int(class_to_color[key][0]), int(class_to_color[key][1]),
int(class_to_color[key][2])), 2)
textLabel = '{}: {}%'.format(key, int(100 * new_probs[jk]))
all_dets.append((key, 100 * new_probs[jk]))
(retval, baseLine) = cv2.getTextSize(textLabel,
cv2.FONT_HERSHEY_COMPLEX,
1, 1)
# To avoid putting text outside the frame
# replace the legende if the box is outside the image
if real_y1 < 20 and real_y2 < img.shape[0]:
textOrg = (real_x1, real_y2 + 5)
elif real_y1 < 20 and real_y2 > img.shape[0]:
textOrg = (real_x1, img.shape[0] - 10)
else:
textOrg = (real_x1, real_y1 + 5)
cv2.rectangle(
img, (textOrg[0] - 5, textOrg[1] + baseLine - 5),
(textOrg[0] + retval[0] + 5, textOrg[1] - retval[1] - 5),
(0, 0, 0), 2)
cv2.rectangle(
img, (textOrg[0] - 5, textOrg[1] + baseLine - 5),
(textOrg[0] + retval[0] + 5, textOrg[1] - retval[1] - 5),
(255, 255, 255), -1)
cv2.putText(img, textLabel, textOrg, cv2.FONT_HERSHEY_DUPLEX,
1, (0, 0, 0), 1)
if print_dets:
print(all_dets)
if export:
return boxes_export
else:
return img
.format(mean_overlapping_bboxes, epoch_length))
if mean_overlapping_bboxes == 0:
print(
'RPN is not producing bounding boxes that overlap the ground truth boxes. Check RPN settings or keep training.'
)
X, Y, img_data = data_gen_train.next()
loss_rpn = model_rpn.train_on_batch(X, Y)
P_rpn = model_rpn.predict_on_batch(X)
R = roi_helpers.rpn_to_roi(P_rpn[0],
P_rpn[1],
C,
K.image_dim_ordering(),
use_regr=True,
overlap_thresh=0.7,
max_boxes=300)
# note: calc_iou converts from (x1,y1,x2,y2) to (x,y,w,h) format
X2, Y1, Y2 = roi_helpers.calc_iou(R, img_data, C,
class_mapping)
if X2 is None:
rpn_accuracy_rpn_monitor.append(0)
rpn_accuracy_for_epoch.append(0)
continue
neg_samples = np.where(Y1[0, :, -1] == 1)
pos_samples = np.where(Y1[0, :, -1] == 0)