anchor_scales = [3, 4, 5]
# BBOX OP
# INV inv_blobs_op OP = return to the original Coordinate
# INV target_inv_blobs_op OP = return to the original Coordinate (indices )
inv_blobs_op , target_inv_blobs_op = inv_transform_layer(rpn_bbox_pred , cfg_key = phase_train , \
_feat_stride = _feat_stride , anchor_scales =anchor_scales , indices = indice_op)
blobs_op, scores_op = roi.roi_proposal(rpn_cls,
rpn_bbox_pred,
im_dims,
_feat_stride,
anchor_scales,
is_training=True)
cost_op = rpn_cls_loss_op + rpn_bbox_loss_op
train_cls_op = optimizer(rpn_cls_loss_op, lr=0.01)
train_bbox_op = optimizer(rpn_bbox_loss_op, lr=0.001)
train_op = optimizer(cost_op, lr=0.001)
sess = sess_start()
max_iter = 55000 * 100
for i in range(2, max_iter):
src_img, src_gt_boxes = next_img_gtboxes(i)
h, w = np.shape(src_img)
src_im_dims = [(h, w)]
rpn_cls_score = np.zeros(
[1, int(math.ceil(h / 8.)),
int(math.ceil(w / 8.)), 512])
rpn_labels, rpn_bbox_targets, rpn_bbox_inside_weights, rpn_bbox_outside_weights, bbox_targets, bbox_inside_weights, bbox_outside_weights = anchor_target(
rpn_cls_score=rpn_cls_score,
rpn_bbox_pred = rpn_bbox_layer(top_conv)
# CLS LOSS
rpn_cls_loss_op ,A_op ,B_op = rpn_cls_loss(rpn_cls , rpn_labels_op)
# BBOX LOSS
# C_op : indiced rpn bbox pred op
# D_op :indiced rpn target op
# E_op : rpn_inside_weights
# F_op : indices
rpn_bbox_loss_op , diff_op , C_op , D_op ,E_op ,F_op = \
rpn_bbox_loss(rpn_bbox_pred ,bbox_targets_op , bbox_inside_weights_op , bbox_outside_weights_op , rpn_labels_op)
cost_op = rpn_bbox_loss_op + rpn_cls_loss_op
rpn_cls_train_op = optimizer(rpn_cls_loss_op)
rpn_bbox_train_op = optimizer(rpn_bbox_loss_op )
train_op = optimizer(rpn_bbox_loss_op )
sess=sess_start()
for i in range(2,55000 * 100):
src_img , src_gt_boxes =next_img_gtboxes(i)
h,w=np.shape(src_img)
src_im_dims = [(h,w)]
anchor_scales = [3,4,5]
rpn_cls_score=np.zeros([1,int(math.ceil(h/8.)),int(math.ceil(w/8.)),512])
rpn_labels, rpn_bbox_targets, rpn_bbox_inside_weights, rpn_bbox_outside_weights, bbox_targets, bbox_inside_weights, bbox_outside_weights = anchor_target(
rpn_cls_score=rpn_cls_score, gt_boxes=src_gt_boxes, im_dims=src_im_dims, _feat_stride=8, anchor_scales=anchor_scales)
"""
# fast rcnn blobs op shape : 1, N , 4 *n_classes
#
itr_fr_blobs_op = inv_transform_layer_fastrcnn(rois_op, itr_fr_bbox_target_op)
# inv_ptl_rois_op = inv_transform_layer_fastrcnn(ptl_rois_op , ptl_bbox_targets_op)
# 임시로 되돌려 만들어 본다 , ptl 은 추천된 박스 영상과 동일해야 한다.
# FAST RCNN COST
fr_cls_loss_op = fast_rcnn_cls_loss(fast_rcnn_cls_logits, ptl_labels_op)
fr_bbox_loss_op = fast_rcnn_bbox_loss(fast_rcnn_bbox_logits,
ptl_bbox_targets_op,
ptl_bbox_inside_weights_op,
ptl_bbox_outside_weights_op)
# Train op and Loss op
rpn_cost_op = rpn_cls_loss_op + rpn_bbox_loss_op
fr_cost_op = fr_cls_loss_op + fr_bbox_loss_op
cost_op = rpn_cost_op + fr_cost_op
train_op = optimizer(cost_op, lr=0.0001)
# Start Session
sess = sess_start()
# Saver
saver = tf.train.Saver(max_to_keep=10)
# Write log
tb_writer = tf.summary.FileWriter('logs')
tb_writer.add_graph(tf.get_default_graph())
#
max_iter = 485 * 3000
max_acc = 0
# start Training
start_time = time.time()
for i in range(0, max_iter):
try: