def _fcn16s_logits(ftmps, hparams, weights, layers, num_classes, logits_shape,
init_weights):
pool4_proj = project_layer(layers["pool4"],
hparams=hparams,
weights=weights,
out_depth=num_classes,
name="pool4_proj",
init_weights=init_weights)
out_up2x = upsample_layer(ftmps,
hparams,
weights=weights,
upsample_factor=2,
output_shape=tf.shape(pool4_proj),
name="out_up2x",
init_weights=init_weights)
fused_pool4 = tf.add(pool4_proj, out_up2x, "fused_pool4")
fused_pool4_up16x = upsample_layer(fused_pool4,
hparams,
weights=weights,
upsample_factor=16,
output_shape=logits_shape,
name="fused_pool4_up16x",
init_weights=init_weights)
return fused_pool4_up16x
def _fcn8s_logits(ftmps, hparams, weights, layers, num_classes, logits_shape,
init_weights):
pool4 = layers["pool4"]
if hparams.fcn8s_oneshot == True:
pool4 = tf.multiply(pool4, 0.01)
pool4_proj = project_layer(pool4,
hparams=hparams,
weights=weights,
out_depth=num_classes,
name="pool4_proj",
init_weights=init_weights)
pool3 = layers["pool3"]
if hparams.fcn8s_oneshot == True:
pool3 = tf.multiply(pool3, 0.0001)
pool3_proj = project_layer(pool3,
hparams=hparams,
weights=weights,
out_depth=num_classes,
name="pool3_proj",
init_weights=init_weights)
out_up2x = upsample_layer(ftmps,
hparams,
weights=weights,
upsample_factor=2,
output_shape=tf.shape(pool4_proj),
name="out_up2x",
init_weights=init_weights)
fused_pool4 = tf.add(pool4_proj, out_up2x, "fused_pool4")
fused_pool4_up2x = upsample_layer(fused_pool4,
hparams,
weights=weights,
upsample_factor=2,
output_shape=tf.shape(pool3_proj),
name="fused_pool4_up2x",
init_weights=init_weights)
fused_pool3 = tf.add(pool3_proj, fused_pool4_up2x, "fused_pool3")
fused_pool3_up8x = upsample_layer(fused_pool3,
hparams,
weights=weights,
upsample_factor=8,
output_shape=logits_shape,
name="fused_pool3_up8x",
init_weights=init_weights)
return fused_pool3_up8x
def yolo_v3(inputs):
# Darknet
route_1, route_2, inputs = darknet.darknet(inputs)
# Large detections
inputs = layers.conv_layer(name="conv_00", inputs=inputs, filters=512, kernel_size=1)
inputs = layers.conv_layer(name="conv_01", inputs=inputs, filters=1024, kernel_size=3)
inputs = layers.conv_layer(name="conv_02", inputs=inputs, filters=512, kernel_size=1)
inputs = layers.conv_layer(name="conv_03", inputs=inputs, filters=1024, kernel_size=3)
inputs = layers.conv_layer(name="conv_04", inputs=inputs, filters=512, kernel_size=1)
large_detections = layers.conv_layer(name="conv_05", inputs=inputs, filters=1024, kernel_size=3)
large_detections = layers.conv_layer(name="conv_06", inputs=large_detections, filters=3*(NUM_CLASSES+5), kernel_size=1, batch_normalize=False, activation=False)
inputs = layers.conv_layer(name="conv_07", inputs=inputs, filters=256, kernel_size=1)
inputs = layers.upsample_layer(name="upsample_0", inputs=inputs)
inputs = layers.route_layer(name="route_0", inputs=inputs, route=route_2)
# Medium detections
inputs = layers.conv_layer(name="conv_08", inputs=inputs, filters=256, kernel_size=1)
inputs = layers.conv_layer(name="conv_09", inputs=inputs, filters=512, kernel_size=3)
inputs = layers.conv_layer(name="conv_10", inputs=inputs, filters=256, kernel_size=1)
inputs = layers.conv_layer(name="conv_11", inputs=inputs, filters=512, kernel_size=3)
inputs = layers.conv_layer(name="conv_12", inputs=inputs, filters=256, kernel_size=1)
medium_detections = layers.conv_layer(name="conv_13", inputs=inputs, filters=512, kernel_size=3)
medium_detections = layers.conv_layer(name="conv_14", inputs=medium_detections, filters=3*(NUM_CLASSES+5), kernel_size=1, batch_normalize=False, activation=False)
inputs = layers.conv_layer(name="conv_15", inputs=inputs, filters=128, kernel_size=1)
inputs = layers.upsample_layer(name="upsample_1", inputs=inputs)
inputs = layers.route_layer(name="route_1", inputs=inputs, route=route_1)
# Small detections
inputs = layers.conv_layer(name="conv_16", inputs=inputs, filters=128, kernel_size=1)
inputs = layers.conv_layer(name="conv_17", inputs=inputs, filters=256, kernel_size=3)
inputs = layers.conv_layer(name="conv_18", inputs=inputs, filters=128, kernel_size=1)
inputs = layers.conv_layer(name="conv_19", inputs=inputs, filters=256, kernel_size=3)
inputs = layers.conv_layer(name="conv_20", inputs=inputs, filters=128, kernel_size=1)
small_detections = layers.conv_layer(name="conv_21", inputs=inputs, filters=256, kernel_size=3)
small_detections = layers.conv_layer(name="conv_22", inputs=small_detections, filters=3*(NUM_CLASSES+5), kernel_size=1, batch_normalize=False, activation=False)
return [small_detections, medium_detections, large_detections]
def _fcn32s_logits(ftmps, hparams, weights, layers, num_classes, logits_shape,
init_weights):
out_up32x = upsample_layer(ftmps,
hparams,
weights=weights,
upsample_factor=32,
output_shape=logits_shape,
name="out_up32x",
init_weights=init_weights)
return out_up32x
def Yolov3_Tiny(inputs, cfg_file="../cfg/yolov3-depth-tiny.cfg"):
blocks = parse_cfg(cfg_file)
# print(blocks)
x, layers, outputs = inputs, [], []
weights_ptr = 0
config = {}
input_dims = 416
for i, block in enumerate(blocks):
block_type = block["type"]
# print("{} {}".format(i-1, block_type))
# print("Input shape: ", x.shape)
if block_type == "convolutional":
x, layers, weights_ptr = conv_layer(x, block, layers, weights_ptr)
elif block_type == "maxpool":
x, layers = maxpool_layer(x, block, layers)
elif block_type == "upsample":
x, layers = upsample_layer(x, block, layers)
elif block_type == "route":
x, layers = route_layer(x, block, layers)
elif block_type == "yolo":
x, layers, outputs = yolo_layer(x, block, layers, outputs,
input_dims)
elif block_type == "convolutional_transpose":
x, layers = conv_transpose_layer(x, block, layers)
elif block_type == "residual":
x, layers = residual_layer(x, block, layers)
# print("Output shape: ", x.shape)
# print("")
# output_layers = [layers[i - 1] for i in range(len(layers)) if layers[i] is None]
outputs = tf.keras.layers.Concatenate(axis=1)(outputs)
# x = tf.sigmoid(x)
# print(x.shape)
# Run NMS
# outputs = non_maximum_suppression(outputs, confidence=0.5, num_classes=80, nms_threshold=0.5)
return (outputs, x) # outputs: yolo bounding box, x: depth