ishape = [1024, 1024, 3]
feature_map_size = [32, 32]
frame_mode = True
classes = ['face', 'none']
mapping = {0: 0}
iou_thresholds = [0.3, 0.5]
num_of_samples = 64
img_ids = None
ann_file = '../datasets/coco/annotations/instances_face.json'
img_dir = '../datasets/coco/images/face'
coco = COCO(ann_file)
if img_ids is None:
gen = genx(
coco=coco,
img_dir=img_dir,
classes=classes,
limit=[start_example_index, start_example_index+num_of_examples],
ishape=ishape)
else:
gen = genx_selected(
coco=coco,
img_dir=img_dir,
img_ids=img_ids,
ishape=ishape)
for sample_order in range(num_of_examples):
# generate x
x, img_id = next(gen)
# get labels
unified_roi_size=unified_roi_size,
rpn_head_dim=rpn_head_dim,
fc_denses=fc_denses,
block_settings=block_settings,
base_block_trainable=True)
# load weihts
# detection_model.load_weights('{}/detection_weights.h5'.format(output_path))
if True:
for epoch in range(1):
gen = genx(coco=coco,
img_dir=img_dir,
classes=classes,
limit=(start_example_index, start_example_index +
num_of_train_examples + num_of_validation_examples),
ishape=ishape)
#------------------------------
# TRAIN
#------------------------------
print('{}: TRAIN {}'.format(datetime.now().time(), epoch), end='\n')
loss = np.zeros((num_of_train_examples, 1))
for batch_idx in range(num_of_train_examples):
# generate x
x, img_id = next(gen)
resnet_settings = [[32, 32, 128], [3, [1, 1]], [4, [2, 2]], [6, [2, 2]], [3, [2, 2]]]
input_tensor = Input(shape=ishape)
C2, C3, C4, C5 = resnet(
input_tensor=input_tensor,
block_settings=resnet_settings,
use_bias=use_bias,
weight_decay=weight_decay,
trainable=trainable,
bn_trainable=bn_trainable)
model = Model(inputs=input_tensor, outputs=[C2, C3, C4, C5])
model.load_weights('{}/_weights.h5'.format(output_path), by_name=True)
gen = genx(
anno_file_path=anno_file_path,
image_dir=image_dir,
ishape=ishape)
for _ in range(total_examples):
x, _ = next(gen)
prediction = model.predict_on_batch(x)
x = x[0]
c2_3dtensor = prediction[0][0]
c3_3dtensor = prediction[1][0]
c4_3dtensor = prediction[2][0]
c5_3dtensor = prediction[3][0]
c2_2dtensor = tf.math.reduce_mean(input_tensor=c2_3dtensor, axis=2)
c3_2dtensor = tf.math.reduce_mean(input_tensor=c3_3dtensor, axis=2)
c4_2dtensor = tf.math.reduce_mean(input_tensor=c4_3dtensor, axis=2)
abox_4dtensor = tf.constant(value=abox4d, dtype='float32')
abox_2dtensor = tf.reshape(tensor=abox_4dtensor, shape=[-1, 4])
model = build_infer_model(ishape=ishape,
resnet_settings=resnet_settings,
k=len(asizes),
total_classes=total_classes,
abox_2dtensor=abox_2dtensor,
nsm_iou_threshold=nsm_iou_threshold,
nsm_score_threshold=nsm_score_threshold,
nsm_max_output_size=nsm_max_output_size)
model.summary()
# model.load_weights('{}/weights.h5'.format(output_path), by_name=True)
gen = genx(anno_file_path=test_anno_file_path,
image_dir=test_image_dir,
ishape=ishape,
mode='test')
for batch in range(total_test_examples):
batchx_4dtensor, _, image_id = next(gen)
print('{}: 1'.format(datetime.now().time()), end='\n')
boxclz_2dtensor, valid_outputs = model.predict_on_batch(
batchx_4dtensor) # (h*w*k, total_classes+1+4)
print('{}: 2'.format(datetime.now().time()), end='\n')
image = batchx_4dtensor[0]
_, ax = plt.subplots(figsize=(15, 7.35))
ax.imshow(image)
ax.set_xlabel('Image ID: {}'.format(image_id))
for i in range(valid_outputs[0]):
from matplotlib.patches import Rectangle
from utils import box2frame
from datagen import genx, gety
np.set_printoptions(threshold=np.inf, linewidth=np.inf)
ishape = [1024, 1024, 3]
frame_mode = True
mapping = {0: 0}
classes = ['face', 'none']ann_file = '../datasets/coco/annotations/instances_face.json'
img_dir = '../datasets/coco/images/face'
coco = COCO(ann_file)
# cat_ids = coco.getCatIds(catNms=classes)
gen = genx(coco=coco, img_dir=img_dir, classes=classes, limit=[0, 100], ishape=ishape)
for i in range(100):
# generate x
x, img_id = next(gen)
bbox2d, _ = gety(coco, img_id, classes, frame_mode=frame_mode, mapping=mapping)
fig, ax = plt.subplots(figsize=(15, 7.35))
ax.imshow(x/255)
for bbox in bbox2d:
frame = box2frame(box=bbox, apoint=[0, 0])
ax.add_patch(Rectangle(
(frame[0], frame[1]), frame[2], frame[3],
linewidth=1,
