# evaluation model with pascal/voc measures
from __future__ import absolute_import, division, print_function
import os
from imdb import Imdb
from network import Network
data_dir = os.path.join(os.getcwd(), 'data')
anno_dir = os.path.join(data_dir, 'eval_annotation')
images_dir = os.path.join(data_dir, 'images')
imdb = Imdb(anno_dir, images_dir, batch_size=1)
net = Network(is_training=False)
for images, gt_boxes, gt_cls in imdb.next_batch(): # batch_size is 1
box_pred, cls_inds, scores = net.predict(images)
args = parser.parse_args()
imdb = Imdb(anno_dir, images_dir,
batch_size=args.batch_size)
net = Network(is_training=True, lr=args.learn_rate)
train_t = 0
step = 0
print('start training')
for epoch in range(1, args.num_epochs + 1):
epoch_t = time.time()
for images, gt_boxes, gt_cls in imdb.next_batch():
step, bbox_loss, iou_loss, cls_loss = net.fit(images, gt_boxes, gt_cls)
if step % 100 == 0:
print('step: {0} - bbox_loss: {1} - iou_loss: {2} - cls_loss: {3}'.format(
step, bbox_loss, iou_loss, cls_loss))
epoch_t_dif = time.time() - epoch_t
train_t += epoch_t_dif
print('epoch: {0} - time: {1}'.format(epoch,
str(timedelta(seconds=epoch_t_dif))))
if epoch % 5 == 0:
net.save_ckpt(step)