def extract_batch(dataset, config):
with tf.device("/cpu:0"):
bboxer = PriorBoxGrid(config)
data_provider = slim.dataset_data_provider.DatasetDataProvider(
dataset,
num_readers=2,
common_queue_capacity=512,
common_queue_min=32)
if args.segment:
im, bbox, gt, seg = data_provider.get(
['image', 'object/bbox', 'object/label', 'image/segmentation'])
else:
im, bbox, gt = data_provider.get(
['image', 'object/bbox', 'object/label'])
seg = tf.expand_dims(tf.zeros(tf.shape(im)[:2]), 2)
im = tf.to_float(im) / 255
bbox = yxyx_to_xywh(tf.clip_by_value(bbox, 0.0, 1.0))
im, bbox, gt, seg = data_augmentation(im, bbox, gt, seg, config)
inds, cats, refine = bboxer.encode_gt_tf(bbox, gt)
return tf.train.shuffle_batch([im, inds, refine, cats, seg],
args.batch_size,
2048,
64,
num_threads=4)
def extract_batch(dataset, config):
with tf.device("/cpu:0"):
bboxer = PriorBoxGrid(config)
data_provider = slim.dataset_data_provider.DatasetDataProvider(
dataset,
num_readers=2,
common_queue_capacity=512,
common_queue_min=32)
if args.instance and args.segment:
im, bbox, gt, seg, ins, im_h, im_w = \
data_provider.get(['image', 'object/bbox', 'object/label',
'image/segmentation',
'image/instance',
'image/height',
'image/width'])
print("ins: ", ins)
print("bbox: ", bbox)
print("gt: ", gt)
ins_shape = tf.stack([im_h, im_w, args.instance_num], axis=0)
ins = tf.reshape(ins, ins_shape)
elif args.segment:
im, bbox, gt, seg = data_provider.get(
['image', 'object/bbox', 'object/label', 'image/segmentation'])
elif args.instance:
im, bbox, gt, ins, im_h, im_w = \
data_provider.get(['image', 'object/bbox', 'object/label',
'image/instance',
'image/height',
'image/width'])
ins_shape = tf.stack([im_h, im_w, args.instance_num], axis=0)
ins = tf.reshape(ins, ins_shape)
else:
im, bbox, gt = data_provider.get(
['image', 'object/bbox', 'object/label'])
seg = tf.expand_dims(tf.zeros(tf.shape(im)[:2]), 2)
ins = tf.expand_dims(tf.zeros(tf.shape(im)[:2]), 2)
im = tf.to_float(im) / 255
bbox = yxyx_to_xywh(tf.clip_by_value(bbox, 0.0, 1.0))
im, bbox, gt, seg, ins = data_augmentation(im, bbox, gt, seg, ins,
config)
inds, cats, refine, gt_matches = bboxer.encode_gt_tf(bbox, gt)
bbox_pads = args.instance_num - tf.shape(bbox)[0]
paddings = [[0, bbox_pads], [0, 0]]
bbox = tf.pad(bbox, paddings)
bbox = tf.reshape(bbox, (args.instance_num, 4))
return tf.train.shuffle_batch(
[im, inds, refine, cats, seg, ins, gt_matches, bbox],
args.batch_size,
2048,
64,
num_threads=4)
def __init__(self,
sess,
net,
loader,
config,
no_gt=False,
folder=None,
profiler=None):
self.sess = sess
self.net = net
self.loader = loader
self.config = config
self.fm_sizes = self.config['fm_sizes']
self.no_gt = no_gt
self.bboxer = PriorBoxGrid(self.config)
self.profiler = profiler
self.run_options = tf.RunOptions(trace_level=tf.RunOptions.FULL_TRACE)
self.run_metadata = tf.RunMetadata()
self.build_detector()
if folder is not None:
self.directory = folder
else:
self.directory = os.path.join(RESULTS_DIR, args.run_name)
if not os.path.exists(self.directory):
os.makedirs(self.directory)
def extract_batch(dataset, config):
with tf.device("/cpu:0"):
bboxer = PriorBoxGrid(config)
data_provider = slim.dataset_data_provider.DatasetDataProvider(
dataset, num_readers=2,
common_queue_capacity=512, common_queue_min=32)
if args.segment:
im, bbox, gt, seg = data_provider.get(['image', 'object/bbox', 'object/label',
'image/segmentation'])
else:
im, bbox, gt = data_provider.get(['image', 'object/bbox', 'object/label'])
seg = tf.expand_dims(tf.zeros(tf.shape(im)[:2]), 2)
im = tf.to_float(im)/255
bbox = yxyx_to_xywh(tf.clip_by_value(bbox, 0.0, 1.0))
im, bbox, gt, seg = data_augmentation(im, bbox, gt, seg, config)
inds, cats, refine = bboxer.encode_gt_tf(bbox, gt)
return tf.train.shuffle_batch([im, inds, refine, cats, seg],
args.batch_size, 2048, 64, num_threads=4)
def __init__(self, sess, net, loader, config, no_gt=False):
self.sess = sess
self.net = net
self.loader = loader
self.config = config
self.fm_sizes = self.config['fm_sizes']
self.no_gt = no_gt
self.bboxer = PriorBoxGrid(self.config)
self.build_detector()
self.directory = os.path.join(RESULTS_DIR, args.run_name)
if not os.path.exists(self.directory):
os.makedirs(self.directory)
def extract_batch(dataset, config):
with tf.device("/cpu:0"):
bboxer = PriorBoxGrid(config)
data_provider = slim.dataset_data_provider.DatasetDataProvider(
dataset,
num_readers=2,
common_queue_capacity=512,
common_queue_min=32)
if args.instance and args.segment:
im, bbox, gt, seg, ins, im_h, im_w = \
data_provider.get(['image', 'object/bbox', 'object/label',
'image/segmentation',
'image/instance',
'image/height',
'image/width'])
ins_num = tf.cast(tf.shape(bbox)[0], tf.int64)
ins_shape = tf.stack([ins_num, im_h, im_w], axis=0)
ins = tf.reshape(ins, ins_shape)
ins = tf.transpose(ins, [1, 2, 0])
pads = args.instance_num - tf.shape(ins)[-1]
ins = tf.pad(ins, [[0, 0], [0, 0], [0, pads]])
# ins_summary = tf.image.resize_images(ins, (300, 300))
# ins_summary = tf.transpose(ins_summary, [2, 0, 1])
# ins_summary = ins_summary[0]
# ins_summary = tf.reshape(ins_summary, [300, 300])
elif args.segment:
im, bbox, gt, seg = data_provider.get(
['image', 'object/bbox', 'object/label', 'image/segmentation'])
elif args.instance:
im, bbox, gt, ins, im_h, im_w = \
data_provider.get(['image', 'object/bbox', 'object/label',
'image/instance',
'image/height',
'image/width'])
ins_num = tf.cast(tf.shape(bbox)[0], tf.int64)
ins_shape = tf.stack([ins_num, im_h, im_w], axis=0)
ins = tf.reshape(ins, ins_shape)
ins = tf.transpose(ins, [1, 2, 0])
pads = args.instance_num - tf.shape(ins)[-1]
ins = tf.pad(ins, [[0, 0], [0, 0], [0, pads]])
seg = tf.expand_dims(tf.zeros(tf.shape(im)[:2]), 2)
else:
im, bbox, gt = data_provider.get(
['image', 'object/bbox', 'object/label'])
seg = tf.expand_dims(tf.zeros(tf.shape(im)[:2]), 2)
ins = tf.expand_dims(tf.zeros(tf.shape(im)[:2]), 2)
im = tf.to_float(im) / 255
bbox = yxyx_to_xywh(tf.clip_by_value(bbox, 0.0, 1.0))
im, bbox, gt, seg, ins = data_augmentation(im, bbox, gt, seg, ins,
config)
inds, cats, refine, gt_matches = bboxer.encode_gt_tf(bbox, gt)
# ins_summary = ins
not_empty_gt_matches = tf.reduce_sum(gt_matches) > 0
gt_matches_idx = tf.reshape(tf.where(gt_matches > 0), (-1, ))
# gt_matches_idx, _ = tf.unique(gt_matches_idx)
match_roi_info = tf.cond(
not_empty_gt_matches,
true_fn=lambda: get_match_roi_info(gt_matches_idx, roi_info),
false_fn=lambda: tf.zeros((args.instance_num, 5)))
gt_matches = tf.gather(gt_matches, gt_matches_idx)
gt_matches = tf.cond(not_empty_gt_matches, lambda: gt_matches - 1,
lambda: gt_matches)
bbox_matches = tf.gather(bbox, gt_matches)
need_pad = args.instance_num > tf.shape(bbox_matches)[0]
pads = tf.cond(need_pad,
lambda: args.instance_num - tf.shape(bbox_matches)[0],
lambda: 0)
bbox_matches = tf.pad(bbox_matches, [[0, pads], [0, 0]])
bbox_matches = bbox_matches[:args.instance_num, :]
# ins = tf.gather(ins, gt_matches)
# ins = tf.pad(ins, [[0, pads], [0, 0], [0, 0]])
# ins = ins[:args.instance_num, ...]
# if tf.reduce_sum(gt_matches) > 0:
# gt_matches_idx = tf.where(gt_matches > 0)
# match_roi_info = tf.squeeze(tf.gather(roi_info, gt_matches_idx // 6)) # (?, 5)
# # print("gt_matches_idx: ", gt_matches_idx)
# # print("roi_info: ", roi_info)
# print("match roi info shape: ", match_roi_info.shape)
# roi_pads = args.instance_num - tf.shape(match_roi_info)[0]
# tf.pad(match_roi_info, [[0, roi_pads], [0, 0]],
# mode='CONSTANT', constant_values=0)
# match_roi_info = tf.reshape(match_roi_info, (args.instance_num, 5))
# else:
# match_roi_info = tf.zeros((args.instance_num, 5))
need_pad = args.instance_num > tf.shape(gt_matches)[0]
pads = tf.cond(need_pad,
lambda: args.instance_num - tf.shape(gt_matches)[0],
lambda: 0)
gt_matches = tf.pad(gt_matches, [[0, pads]])
gt_matches = gt_matches[:args.instance_num]
ins = extract_matched_gt_instance(bbox_matches, ins,
[args.ins_shape, args.ins_shape],
gt_matches)
ins = tf.reshape(ins,
(args.instance_num, args.ins_shape, args.ins_shape))
tf.summary.tensor_summary('tensor/bbox_matches', bbox_matches)
tf.summary.tensor_summary('tensor/gt_matches', gt_matches)
ins_cats = tf.gather(cats, gt_matches_idx)
need_pad = args.instance_num > tf.shape(ins_cats)[0]
pads = tf.cond(need_pad,
lambda: args.instance_num - tf.shape(ins_cats)[0],
lambda: 0)
ins_cats = tf.pad(ins_cats, [[0, pads]])
ins_cats = ins_cats[:args.instance_num]
ins_cats = tf.reshape(ins_cats, (args.instance_num, ))
return tf.train.shuffle_batch(
[im, inds, refine, cats, ins_cats, seg, ins, match_roi_info],
args.batch_size,
2048,
64,
num_threads=4)
def main(argv=None): # pylint: disable=unused-argument
net = ResNet
depth = 50
loader = VOCLoader('07', 'test')
net = net(config=net_config, depth=depth, training=False)
num_classes = 21
batch_size = args.batch_size
img_size = args.image_size
image_ph = tf.placeholder(shape=[1, img_size, img_size, 3],
dtype=tf.float32,
name='img_ph')
net.create_trunk(image_ph)
bboxer = PriorBoxGrid(net_config)
net.create_multibox_head(num_classes)
confidence = tf.nn.softmax(tf.squeeze(net.outputs['confidence']))
location = tf.squeeze(net.outputs['location'])
good_bboxes = decode_bboxes(location, bboxer.tiling)
detection_list = []
score_list = []
for i in range(1, num_classes):
class_mask = tf.greater(confidence[:, i], args.conf_thresh)
class_scores = tf.boolean_mask(confidence[:, i], class_mask)
class_bboxes = tf.boolean_mask(good_bboxes, class_mask)
K = tf.minimum(tf.size(class_scores), args.top_k_nms)
_, top_k_inds = tf.nn.top_k(class_scores, K)
top_class_scores = tf.gather(class_scores, top_k_inds)
top_class_bboxes = tf.gather(class_bboxes, top_k_inds)
final_inds = tf.image.non_max_suppression(
top_class_bboxes,
top_class_scores,
max_output_size=50,
iou_threshold=args.nms_thresh)
final_class_bboxes = tf.gather(top_class_bboxes, final_inds)
final_scores = tf.gather(top_class_scores, final_inds)
detection_list.append(final_class_bboxes)
score_list.append(final_scores)
net.create_segmentation_head(num_classes)
segmentation = tf.cast(
tf.argmax(tf.squeeze(net.outputs['segmentation']), axis=-1), tf.int32)
times = []
# im = cv2.imread("/home/lear/kshmelko/testImage.jpg")
# im = cv2.cvtColor(im, cv2.COLOR_BGR2RGB)/255.0
# im = im.astype(np.float32)
with tf.Session(config=tf.ConfigProto(allow_soft_placement=True,
log_device_placement=False)) as sess:
sess.run(tf.global_variables_initializer())
ckpt_path = train_dir + '/model.ckpt-%i000' % args.ckpt
log.debug("Restoring checkpoint %s" % ckpt_path)
saver = tf.train.Saver(tf.global_variables())
saver.restore(sess, ckpt_path)
for i in range(200):
im = loader.load_image(loader.get_filenames()[i])
im = cv2.resize(im, (img_size, img_size))
im = im.reshape((1, img_size, img_size, 3))
st = time.time()
sess.run([detection_list, score_list, segmentation],
feed_dict={image_ph: im})
et = time.time()
if i > 10:
times.append(et - st)
m = np.mean(times)
s = np.std(times)
fps = 1 / m
log.info("Mean={0:.2f}ms; Std={1:.2f}ms; FPS={2:.1f}".format(
m * 1000, s * 1000, fps))