Module kerod.layers.anchors
None
None
View Source
import tensorflow as tf
from kerod.core.constants import MAX_IMAGE_DIMENSION
from kerod.utils.documentation import remove_unwanted_doc
__pdoc__ = {}
def generate_anchors(stride: int, scales: tf.Tensor, ratios: tf.Tensor, max_size: int):
"""Will generate a determistic grid based on the input images dimension.
The anchors will be generated once and for all on the biggest possible image.
At each forward according to the shape of the tensor we can extract the corresponding anchors.
Arguments:
stride: Downscaling ratio compared to the original image. At stride 16 your original
image will be 16 times bigger than your actual tensor
scales: The scale of the anchors e.g: 8, 16, 32
ratios: The ratios are the different shapes that you want to apply on your anchors.
e.g: (0.5, 1, 2)
max_size: Maximum size of the input image. The anchors will computed once and for all.
Returns:
A tensor of shape [num_scales * num_ratios * height * width, 4].
The anchors have the format [y_min, x_min, y_max, x_max].
"""
ratios, scales = tf.meshgrid(ratios, scales)
scales = tf.reshape(scales, [-1])
ratios = tf.reshape(ratios, [-1])
ratio_sqrts = tf.sqrt(ratios)
widths = scales / ratio_sqrts
heights = ratios * widths
base_anchors = tf.stack([-widths, -heights, widths, heights], axis=-1) * 0.5
max_stride = tf.math.ceil(max_size / stride)
y_centers = tf.cast(tf.range(max_stride), dtype=scales.dtype) * stride
x_centers = tf.cast(tf.range(max_stride), dtype=scales.dtype) * stride
x_centers, y_centers = tf.meshgrid(x_centers, y_centers)
shifts = tf.stack([x_centers, y_centers, x_centers, y_centers], axis=-1)
anchors = tf.expand_dims(base_anchors, 0) + tf.expand_dims(shifts, 2)
return tf.gather(anchors, [1, 0, 3, 2], axis=-1)
class Anchors(tf.keras.layers.Layer):
"""Will generate a determistic grid and store it in memory to avoid recompute it at each run.
At each forward according to the shape of the tensor we can extract the corresponding anchors.
Arguments:
stride: Downscaling ratio compared to the original image. At stride 16 your original
image will be 16 times bigger than your actual tensor
scales: The scale of the anchors e.g: 8, 16, 32
ratios: The ratios are the different shapes that you want to apply on your anchors.
e.g: (0.5, 1, 2)
max_size: Maximum size of the input image. The anchors will computed once and for all.
Call arguments:
inputs: A tensor of shape [batch_size, height, widht, channel]
Call returns:
tf.Tensor: A tensor of shape [num_scales * num_ratios * height * width, 4].
The anchors have the format [y_min, x_min, y_max, x_max].
"""
def __init__(self, stride, scales, ratios, **kwargs):
super().__init__(**kwargs)
self._stride = stride
self._scales = scales
self._ratios = ratios
self._anchors = generate_anchors(stride,
tf.constant([scales], self._compute_dtype),
tf.constant(ratios, self._compute_dtype),
max_size=MAX_IMAGE_DIMENSION)
def call(self, inputs):
"""Return anchors based on the shape of the input tensors
Arguments:
inputs: A tensor of shape [batch_size, height, widht, channel]
Returns:
tf.Tensor: A tensor of shape [num_scales * num_ratios * height * width, 4].
The anchors have the format [y_min, x_min, y_max, x_max].
"""
shape = tf.shape(inputs)
height, width = shape[1], shape[2]
anchors = self._anchors[:height, :width]
return tf.reshape(anchors, (-1, 4))
def get_config(self):
config = super().get_config()
config['stride'] = self._stride
config['scales'] = self._scales
config['ratios'] = self._ratios
return config
remove_unwanted_doc(Anchors, __pdoc__)
Variables
MAX_IMAGE_DIMENSION
Functions
generate_anchors
def generate_anchors(
stride: int,
scales: tensorflow.python.framework.ops.Tensor,
ratios: tensorflow.python.framework.ops.Tensor,
max_size: int
)
Will generate a determistic grid based on the input images dimension.
The anchors will be generated once and for all on the biggest possible image. At each forward according to the shape of the tensor we can extract the corresponding anchors.
Parameters:
| Name | Description |
|---|---|
| stride | Downscaling ratio compared to the original image. At stride 16 your original image will be 16 times bigger than your actual tensor |
| scales | The scale of the anchors e.g: 8, 16, 32 |
| ratios | The ratios are the different shapes that you want to apply on your anchors. e.g: (0.5, 1, 2) |
| max_size | Maximum size of the input image. The anchors will computed once and for all. |
Returns:
| Type | Description |
|---|---|
| None | A tensor of shape [num_scales * num_ratios * height * width, 4]. The anchors have the format [y_min, x_min, y_max, x_max]. |
View Source
def generate_anchors(stride: int, scales: tf.Tensor, ratios: tf.Tensor, max_size: int):
"""Will generate a determistic grid based on the input images dimension.
The anchors will be generated once and for all on the biggest possible image.
At each forward according to the shape of the tensor we can extract the corresponding anchors.
Arguments:
stride: Downscaling ratio compared to the original image. At stride 16 your original
image will be 16 times bigger than your actual tensor
scales: The scale of the anchors e.g: 8, 16, 32
ratios: The ratios are the different shapes that you want to apply on your anchors.
e.g: (0.5, 1, 2)
max_size: Maximum size of the input image. The anchors will computed once and for all.
Returns:
A tensor of shape [num_scales * num_ratios * height * width, 4].
The anchors have the format [y_min, x_min, y_max, x_max].
"""
ratios, scales = tf.meshgrid(ratios, scales)
scales = tf.reshape(scales, [-1])
ratios = tf.reshape(ratios, [-1])
ratio_sqrts = tf.sqrt(ratios)
widths = scales / ratio_sqrts
heights = ratios * widths
base_anchors = tf.stack([-widths, -heights, widths, heights], axis=-1) * 0.5
max_stride = tf.math.ceil(max_size / stride)
y_centers = tf.cast(tf.range(max_stride), dtype=scales.dtype) * stride
x_centers = tf.cast(tf.range(max_stride), dtype=scales.dtype) * stride
x_centers, y_centers = tf.meshgrid(x_centers, y_centers)
shifts = tf.stack([x_centers, y_centers, x_centers, y_centers], axis=-1)
anchors = tf.expand_dims(base_anchors, 0) + tf.expand_dims(shifts, 2)
return tf.gather(anchors, [1, 0, 3, 2], axis=-1)
Classes
Anchors
class Anchors(
stride,
scales,
ratios,
**kwargs
)
At each forward according to the shape of the tensor we can extract the corresponding anchors.
Arguments
| Name | Description |
|---|---|
| stride | Downscaling ratio compared to the original image. At stride 16 your original image will be 16 times bigger than your actual tensor |
| scales | The scale of the anchors e.g: 8, 16, 32 |
| ratios | The ratios are the different shapes that you want to apply on your anchors. e.g: (0.5, 1, 2) |
| max_size | Maximum size of the input image. The anchors will computed once and for all. |
Call arguments
| Name | Description |
|---|---|
| inputs | A tensor of shape [batch_size, height, widht, channel] |
Call returns
| Type | Description |
|---|---|
| tf.Tensor | A tensor of shape [num_scales * num_ratios * height * width, 4]. The anchors have the format [y_min, x_min, y_max, x_max]. |
Ancestors (in MRO)
- tensorflow.python.keras.engine.base_layer.Layer
- tensorflow.python.module.module.Module
- tensorflow.python.training.tracking.tracking.AutoTrackable
- tensorflow.python.training.tracking.base.Trackable
- tensorflow.python.keras.utils.version_utils.LayerVersionSelector
Methods
call
def call(
self,
inputs
)
Return anchors based on the shape of the input tensors
Parameters:
| Name | Description |
|---|---|
| inputs | A tensor of shape [batch_size, height, widht, channel] |
Returns:
| Type | Description |
|---|---|
| tf.Tensor | A tensor of shape [num_scales * num_ratios * height * width, 4]. The anchors have the format [y_min, x_min, y_max, x_max]. |
View Source
def call(self, inputs):
"""Return anchors based on the shape of the input tensors
Arguments:
inputs: A tensor of shape [batch_size, height, widht, channel]
Returns:
tf.Tensor: A tensor of shape [num_scales * num_ratios * height * width, 4].
The anchors have the format [y_min, x_min, y_max, x_max].
"""
shape = tf.shape(inputs)
height, width = shape[1], shape[2]
anchors = self._anchors[:height, :width]
return tf.reshape(anchors, (-1, 4))