Grayscale prediction

[pmikas]

Hello!

I trained a model in grayscale, so I am also trying to predict in grayscale. Is there a way to set the prediction with 1 channel?

Regards

[hhk7734]

tensorflow-yolov4/py_src/yolov4/tf/__init__.py

Lines 55 to 84 in 4f75970

	def make_model(
	self,
	activation0: str = "mish",
	activation1: str = "leaky",
	kernel_regularizer=tf.keras.regularizers.l2(0.0005),
	):
	# pylint: disable=missing-function-docstring
	self._has_weights = False
	backend.clear_session()
	# height, width, channels
	inputs = layers.Input([self.input_size[1], self.input_size[0], 3])
	if self.tiny:
	self.model = yolov4.YOLOv4Tiny(
	anchors=self.anchors,
	num_classes=len(self.classes),
	xyscales=self.xyscales,
	activation=activation1,
	kernel_regularizer=kernel_regularizer,
	)
	else:
	self.model = yolov4.YOLOv4(
	anchors=self.anchors,
	num_classes=len(self.classes),
	xyscales=self.xyscales,
	activation0=activation0,
	activation1=activation1,
	kernel_regularizer=kernel_regularizer,
	)
	self.model(inputs)

Line 66 - inputs = layers.Input([self.input_size[1], self.input_size[0], 3])
try to change 3 to 1.

[pmikas]

Thanks for the reply

It doesn't work. Even adjusting other parts of the code, the output isn't the same as from darknet

[hhk7734]

I'm working on yolov4 v3.0.
I will try after release.

[hhk7734]

Can you share 1 channel dataset?