#!/usr/bin/env python
# ******************************************************************************
# Copyright 2023 Brainchip Holdings Ltd.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ******************************************************************************
"""
Data generator for CenterNet training
"""
__all__ = ["BatchCenternetGenerator"]
import numpy as np
import tensorflow as tf
from imgaug import augmenters as iaa
from ..detection.batch_generator import BatchGenerator
[docs]class BatchCenternetGenerator(BatchGenerator):
"""
Data generator used for the training process
"""
@staticmethod
def build_aug_pipeline():
# augmentors by https://github.com/aleju/imgaug
def sometimes(aug): return iaa.Sometimes(0.5, aug)
# All augmenters with per_channel=0.5 will sample one value per
# image in 50% of all cases. In all other cases they will sample new
# values per channel.
return iaa.Sequential(
[
# apply the following augmenters to most images
sometimes(iaa.Affine()),
sometimes(iaa.Cutout(nb_iterations=(0, 3), position="uniform", size=(0, 0.3))),
# execute 0 to 5 of the following (less important) augmenters
# per image. Don't execute all of them, as that would often be
# way too strong
iaa.SomeOf(
(0, 5),
[
iaa.OneOf([
# blur images with a sigma between 0 and 3.0
iaa.GaussianBlur((0, 3.0)),
# blur image using local means (kernel sizes between 2 and 7)
iaa.AverageBlur(k=(2, 7)),
# blur image using local medians (kernel sizes between 3 and 11)
iaa.MedianBlur(k=(3, 11)),
]),
# sharpen images
iaa.Sharpen(alpha=(0, 1.0), lightness=(0.75, 1.5)),
iaa.LinearContrast((0.5, 2.0), per_channel=0.5),
iaa.AdditiveGaussianNoise(loc=0, scale=(0.0, 0.05 * 255), per_channel=0.5),
# randomly remove up to 10% of the pixels
iaa.OneOf([iaa.Dropout((0.01, 0.1), per_channel=0.5),
iaa.CoarseDropout((0.01, 0.05), size_percent=0.5)]),
# change brightness of images
iaa.Add((-10, 10), per_channel=0.5),
iaa.Multiply((0.5, 1.5), per_channel=0.5),
iaa.OneOf([
iaa.pillike.Equalize(),
iaa.pillike.Autocontrast()
])
], random_order=True)
],
random_order=True)
def __getitem__(self, idx):
lower_bound = idx * self._batch_size
upper_bound = (idx + 1) * self._batch_size
if upper_bound > self._data_length:
upper_bound = self._data_length
lower_bound = upper_bound - self._batch_size
instance_count = 0
N = upper_bound - lower_bound
x_batch = np.zeros((N, self._input_shape[1], self._input_shape[0], self._input_shape[2]))
y_batch = np.zeros((N, self._grid_size[1], self._grid_size[0], 2 + 2 + len(self._labels)))
for train_instance in self._data[lower_bound:upper_bound]:
# augment input image and fix object's position and size
img, all_objs = self.aug_image(train_instance, jitter=self._jitter)
for obj in all_objs:
if obj['x2'] > obj['x1'] and obj['y2'] > obj['y1'] and obj['label'] in self._labels:
# find center
center_x = .5 * (obj['x1'] + obj['x2'])
# express it in output size
center_x = center_x / (float(self._input_shape[0]) / self._grid_size[0])
center_y = .5 * (obj['y1'] + obj['y2'])
center_y = center_y / (float(self._input_shape[1]) / self._grid_size[0])
grid_x = int(np.floor(center_x))
grid_y = int(np.floor(center_y))
if grid_x < self._grid_size[0] and grid_y < self._grid_size[1]:
obj_indx = self._labels.index(obj['label'])
scale_box_w = (obj['x2'] - obj['x1']) / \
(float(self._input_shape[0]) / self._grid_size[0])
scale_box_h = (obj['y2'] - obj['y1']) / (
float(self._input_shape[1]) / self._grid_size[0])
# get the center point and use a gaussian kernel as the target
radius = self._gaussian_radius([scale_box_h, scale_box_w], min_overlap=0.3)
# check that the radius is positive
radius = np.max(int(radius), 0)
heatmap = np.zeros(
(self._grid_size[0], self._grid_size[1], len(self._labels)))
heatmap = self._gen_gaussian_target(
heatmap, [grid_y, grid_x], obj_indx, radius)
# assign ground truth to y_batch
y_batch[instance_count, ..., :len(self._labels)] += heatmap
y_batch[instance_count, grid_y, grid_x, -4] = scale_box_w
y_batch[instance_count, grid_y, grid_x, -3] = scale_box_h
y_batch[instance_count, grid_y, grid_x, -2] = center_x - grid_x
y_batch[instance_count, grid_y, grid_x, -1] = center_y - grid_y
# assign input image to x_batch
x_batch[instance_count] = img
# increase instance counter in current batch
instance_count += 1
return x_batch, y_batch
@staticmethod
def _gaussian2D(radius, sigma=1, eps=tf.keras.backend.epsilon()):
"""Generate 2D gaussian kernel.
Args:
radius (int): Radius of gaussian kernel.
sigma (int, optional): Sigma of gaussian function. Defaults to 1.
eps (float, optional): Epsilon value. Defaults to 1e-7.
Returns:
np.array: Gaussian kernel with a ``(2 * radius + 1) * (2 * radius + 1)`` shape.
"""
x = np.reshape(np.arange(-radius, radius + 1), [1, -1])
y = np.reshape(np.arange(-radius, radius + 1), [-1, 1])
h = np.exp(-(x * x + y * y) / (2 * sigma * sigma))
# Clamp smaller values to zero
h[h < (eps * h.max())] = 0
return h
@staticmethod
def _gen_gaussian_target(heatmap, center, obj_idx, radius):
"""Generate 2D gaussian heatmap.
Args:
heatmap (np.array): Input heatmap, the gaussian kernel will cover on it and maintain
the max value.
center (list[int]): Coordinates of gaussian kernel's center.
obj_idx (int): The class index for the center point.
radius (int): Radius of gaussian kernel.
Returns:
out_heatmap (np.array): Updated heatmap covered by gaussian kernel.
Note:
Taken from pytorch
"""
diameter = 2 * radius + 1
gaussian_kernel = BatchCenternetGenerator._gaussian2D(radius, sigma=diameter / 6)
x, y = center
height, width = heatmap.shape[:2]
# Find the smallest value so that if the point is near the edge we don't end outside
# (e.g. x = 3 and radius is 10, then we go from the x-3 to x+10)
left = np.min([x, radius])
right = np.min([width - x, radius + 1])
top = np.min([y, radius])
bottom = np.min([height - y, radius + 1])
# Compare the gaussian kernel to the heatmap (in case there's already a point of
# interest there) and keep the max value
flattened_kernel = np.reshape(gaussian_kernel, [-1])
# Range the dimensions
i = 0
for d0 in np.arange(y - top, y + bottom):
for d1 in np.arange(x - left, x + right):
heatmap[d1, d0, obj_idx] = flattened_kernel[i]
i += 1
return heatmap
@staticmethod
def _gaussian_radius(det_size, min_overlap):
r"""Generate 2D gaussian radius.
This function is modified from the `official github repo
<https://github.com/princeton-vl/CornerNet-Lite/blob/master/core/sample/
utils.py#L65>`_.
Given ``min_overlap``, radius could computed by a quadratic equation
according to Vieta's formulas.
There are 3 cases for computing gaussian radius, details are following:
- Case 1: one corner is inside the gt box and the other is outside.
.. code:: text
|< width >|
lt-+----------+ -
| | | ^
+--x----------+--+
| | | |
| | | | height
| | overlap | |
| | | |
| | | | v
+--+---------br--+ -
| | |
+----------+--x
To ensure IoU of generated box and gt box is larger than ``min_overlap``:
.. math::
\cfrac{(w-r)*(h-r)}{w*h+(w+h)r-r^2} \ge {iou} \quad\Rightarrow\quad
{r^2-(w+h)r+\cfrac{1-iou}{1+iou}*w*h} \ge 0 \\
{a} = 1,\quad{b} = {-(w+h)},\quad{c} = {\cfrac{1-iou}{1+iou}*w*h}
{r} \le \cfrac{-b-\sqrt{b^2-4*a*c}}{2*a}
- Case 2: both two corners are inside the gt box.
.. code:: text
|< width >|
lt-+----------+ -
| | | ^
+--x-------+ |
| | | |
| |overlap| | height
| | | |
| +-------x--+
| | | v
+----------+-br -
To ensure IoU of generated box and gt box is larger than ``min_overlap``:
.. math::
\cfrac{(w-2*r)*(h-2*r)}{w*h} \ge {iou} \quad\Rightarrow\quad
{4r^2-2(w+h)r+(1-iou)*w*h} \ge 0 \\
{a} = 4,\quad {b} = {-2(w+h)},\quad {c} = {(1-iou)*w*h}
{r} \le \cfrac{-b-\sqrt{b^2-4*a*c}}{2*a}
- Case 3: both two corners are outside the gt box.
.. code:: text
|< width >|
x--+----------------+
| | |
+-lt-------------+ | -
| | | | ^
| | | |
| | overlap | | height
| | | |
| | | | v
| +------------br--+ -
| | |
+----------------+--x
To ensure IoU of generated box and gt box is larger than ``min_overlap``:
.. math::
\cfrac{w*h}{(w+2*r)*(h+2*r)} \ge {iou} \quad\Rightarrow\quad
{4*iou*r^2+2*iou*(w+h)r+(iou-1)*w*h} \le 0 \\
{a} = {4*iou},\quad {b} = {2*iou*(w+h)},\quad {c} = {(iou-1)*w*h} \\
{r} \le \cfrac{-b+\sqrt{b^2-4*a*c}}{2*a}
Args:
det_size (list[int]): Shape of object.
min_overlap (float): Min IoU with ground truth for boxes generated by
keypoints inside the gaussian kernel.
Returns:
radius (int): Radius of gaussian kernel.
Notes:
Explanation of figure: ``lt`` and ``br`` indicates the left-top and bottom-right
corner of ground truth box. ``x`` indicates the generated corner at the limited
position when ``radius=r``.
"""
height, width = det_size
a1 = 1
b1 = (height + width)
c1 = width * height * (1 - min_overlap) / (1 + min_overlap)
sq1 = np.sqrt(b1**2 - 4 * a1 * c1)
r1 = (b1 - sq1) / (2 * a1)
a2 = 4
b2 = 2 * (height + width)
c2 = (1 - min_overlap) * width * height
sq2 = np.sqrt(b2**2 - 4 * a2 * c2)
r2 = (b2 - sq2) / (2 * a2)
a3 = 4 * min_overlap
b3 = -2 * min_overlap * (height + width)
c3 = (min_overlap - 1) * width * height
sq3 = np.sqrt(b3**2 - 4 * a3 * c3)
r3 = (b3 + sq3) / (2 * a3)
return np.min([r1, r2, r3])