Faster R-CNN
import tensorflow as tf
from skimage.io import imread
im = imread('zebra.jpg')
import matplotlib.pyplot as plt
plt.imshow(im)
<matplotlib.image.AxesImage at 0x17bbc7e20>
import numpy as np
bbs = np.array([[120,25,200,165], [300,50,480,320]])
import matplotlib.patches as pt
fig, ax = plt.subplots()
ax.imshow(im)
for i in bbs:
ax.add_patch(pt.Rectangle((i[0],i[1]),i[2]-i[0],i[3]-i[1], fill=None, color='red'))
# %pip install -U opencv-python
Collecting opencv-python
Downloading opencv_python-4.6.0.66-cp37-abi3-macosx_11_0_arm64.whl (30.0 MB)
[2K [90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━[0m [32m30.0/30.0 MB[0m [31m49.2 MB/s[0m eta [36m0:00:00[0m00:01[0m00:01[0m
[?25hRequirement already satisfied: numpy>=1.17.3 in /Users/hdj/miniforge3/envs/tf/lib/python3.9/site-packages (from opencv-python) (1.22.4)
Installing collected packages: opencv-python
Successfully installed opencv-python-4.6.0.66
Note: you may need to restart the kernel to use updated packages.
import cv2
im2 = cv2.imread('zebra.jpg')
im2 = im2[...,::-1]
im2_ = im2.copy()
for i in bbs:
cv2.rectangle(im2_, (i[0],i[1]), (i[2],i[3]), color=(255,0,0))
plt.imshow(im2_)
<matplotlib.image.AxesImage at 0x17ee52940>
im2.shape
(333, 500, 3)
im3 = cv2.resize(im2, (800,800))
800/333
2.4024024024024024
800/500
1.6
im3.shape
(800, 800, 3)
im2.shape[0]/im3.shape[0]
500
bbs2 = bbs.copy()
bbs2
array([[120, 25, 200, 165],
[300, 50, 480, 320]])
bbs2[:,0] = bbs2[:,0]*im3.shape[1]/im2.shape[1]
bbs2[:,1] = bbs2[:,1]*im3.shape[0]/im2.shape[0]
bbs2[:,2] = bbs2[:,2]*im3.shape[1]/im2.shape[1]
bbs2[:,3] = bbs2[:,3]*im3.shape[0]/im2.shape[0]
bbs2
array([[192, 60, 320, 396],
[480, 120, 768, 768]])
bbs2[:,0] = bbs2[:,0]*800/500
bbs2[:,1] = bbs2[:,1]*800/300
bbs2[:,2] = bbs2[:,2]*800/500
bbs2[:,3] = bbs2[:,3]*800/300
bbs2
array([[192, 66, 320, 440],
[480, 133, 768, 853]])
im3.shape
(800, 800, 3)
im3_ = im3.copy()
for i in bbs2:
cv2.rectangle(im3_,(i[0],i[1]), (i[2],i[3]), color=(255,0,0), thickness=3)
plt.imshow(im3_)
<matplotlib.image.AxesImage at 0x2c9e6be20>
# VGG
vgg = tf.keras.applications.VGG16(include_top=False)
Metal device set to: Apple M1
2022-08-11 20:05:16.587170: I tensorflow/core/common_runtime/pluggable_device/pluggable_device_factory.cc:305] Could not identify NUMA node of platform GPU ID 0, defaulting to 0. Your kernel may not have been built with NUMA support.
2022-08-11 20:05:16.587590: I tensorflow/core/common_runtime/pluggable_device/pluggable_device_factory.cc:271] Created TensorFlow device (/job:localhost/replica:0/task:0/device:GPU:0 with 0 MB memory) -> physical PluggableDevice (device: 0, name: METAL, pci bus id: <undefined>)
vgg.summary()
Model: "vgg16"
_________________________________________________________________
Layer (type) Output Shape Param #
=================================================================
input_1 (InputLayer) [(None, None, None, 3)] 0
block1_conv1 (Conv2D) (None, None, None, 64) 1792
block1_conv2 (Conv2D) (None, None, None, 64) 36928
block1_pool (MaxPooling2D) (None, None, None, 64) 0
block2_conv1 (Conv2D) (None, None, None, 128) 73856
block2_conv2 (Conv2D) (None, None, None, 128) 147584
block2_pool (MaxPooling2D) (None, None, None, 128) 0
block3_conv1 (Conv2D) (None, None, None, 256) 295168
block3_conv2 (Conv2D) (None, None, None, 256) 590080
block3_conv3 (Conv2D) (None, None, None, 256) 590080
block3_pool (MaxPooling2D) (None, None, None, 256) 0
block4_conv1 (Conv2D) (None, None, None, 512) 1180160
block4_conv2 (Conv2D) (None, None, None, 512) 2359808
block4_conv3 (Conv2D) (None, None, None, 512) 2359808
block4_pool (MaxPooling2D) (None, None, None, 512) 0
block5_conv1 (Conv2D) (None, None, None, 512) 2359808
block5_conv2 (Conv2D) (None, None, None, 512) 2359808
block5_conv3 (Conv2D) (None, None, None, 512) 2359808
block5_pool (MaxPooling2D) (None, None, None, 512) 0
=================================================================
Total params: 14,714,688
Trainable params: 14,714,688
Non-trainable params: 0
_________________________________________________________________
len(vgg.layers) # input 1개, MaxPlooing 5개 제외 13 layers
19
im3.shape
(800, 800, 3)
vgg(im3[tf.newaxis])
<tf.Tensor: shape=(1, 25, 25, 512), dtype=float32, numpy=
array([[[[ 0. , 0. , 0. , ..., 0. ,
0. , 0. ],
[ 0. , 0. , 0. , ..., 0. ,
0. , 0. ],
[ 0. , 0. , 0.14235538, ..., 0. ,
0. , 0. ],
...,
[ 0. , 0. , 0.72310966, ..., 0. ,
2.1115327 , 0. ],
[ 0. , 0. , 0. , ..., 0. ,
3.5494702 , 0. ],
[ 0. , 0. , 2.5827112 , ..., 0. ,
1.2787983 , 0. ]],
[[ 0. , 0. , 0. , ..., 0. ,
0. , 0. ],
[ 0. , 0. , 0. , ..., 0. ,
0. , 0. ],
[ 0. , 0. , 0. , ..., 0. ,
0. , 0. ],
...,
[ 0. , 0. , 0. , ..., 0. ,
0.07996958, 0. ],
[ 0. , 0. , 0. , ..., 0. ,
0. , 0. ],
[ 0. , 0. , 0. , ..., 0. ,
0. , 0. ]],
[[ 0. , 0. , 0. , ..., 0. ,
0. , 0. ],
[ 0. , 0. , 0. , ..., 0. ,
0. , 0. ],
[ 0. , 0. , 0. , ..., 0. ,
0. , 0. ],
...,
[ 0. , 0. , 0. , ..., 0. ,
0. , 0. ],
[ 0. , 0. , 0. , ..., 0. ,
0.77511567, 0. ],
[ 0. , 0. , 0. , ..., 0. ,
3.3173277 , 0. ]],
...,
[[ 0. , 0. , 0. , ..., 0. ,
2.6076238 , 0.2472178 ],
[ 0. , 0. , 0. , ..., 0. ,
0. , 1.459591 ],
[ 0. , 0. , 0. , ..., 0. ,
0. , 0.13755503],
...,
[ 0. , 0. , 0. , ..., 0. ,
0. , 0. ],
[ 0. , 0. , 0. , ..., 0. ,
0. , 0. ],
[ 0. , 0. , 0. , ..., 0. ,
5.72634 , 0. ]],
[[15.054876 , 0. , 0. , ..., 0. ,
2.1826699 , 0. ],
[11.117959 , 0. , 0. , ..., 0. ,
0. , 0. ],
[ 0. , 0. , 0. , ..., 0. ,
0. , 0. ],
...,
[ 0. , 0. , 0. , ..., 0. ,
0. , 0. ],
[ 0. , 0. , 0. , ..., 0. ,
0. , 0. ],
[ 0. , 0. , 0. , ..., 0. ,
0. , 0. ]],
[[ 0. , 0. , 0. , ..., 0. ,
0. , 0. ],
[ 0. , 0. , 0. , ..., 0. ,
0.2981035 , 0. ],
[ 0. , 0. , 0. , ..., 0. ,
1.1234553 , 0. ],
...,
[ 0. , 0. , 0. , ..., 0. ,
0. , 0. ],
[ 0. , 0. , 0. , ..., 0. ,
0. , 0. ],
[ 0. , 0. , 0. , ..., 0. ,
0. , 0. ]]]], dtype=float32)>
from tensorflow.keras.backend import function
o = function([vgg.layers[0].input], [vgg.layers[0].output])
o(im3)[0].shape
(800, 800, 3)
for l in vgg.layers:
o = function([vgg.layers[0].input], [l.output])
print(o(im3[tf.newaxis])[0].shape)
(1, 800, 800, 3)
(1, 800, 800, 64)
(1, 800, 800, 64)
(1, 400, 400, 64)
(1, 400, 400, 128)
(1, 400, 400, 128)
(1, 200, 200, 128)
(1, 200, 200, 256)
(1, 200, 200, 256)
(1, 200, 200, 256)
(1, 100, 100, 256)
(1, 100, 100, 512)
(1, 100, 100, 512)
(1, 100, 100, 512)
(1, 50, 50, 512)
(1, 50, 50, 512)
(1, 50, 50, 512)
(1, 50, 50, 512)
(1, 25, 25, 512)
model = tf.keras.Model(vgg.layers[0].input, vgg.layers[-2].output)
model.summary()
Model: "model_22"
_________________________________________________________________
Layer (type) Output Shape Param #
=================================================================
input_1 (InputLayer) [(None, None, None, 3)] 0
block1_conv1 (Conv2D) (None, None, None, 64) 1792
block1_conv2 (Conv2D) (None, None, None, 64) 36928
block1_pool (MaxPooling2D) (None, None, None, 64) 0
block2_conv1 (Conv2D) (None, None, None, 128) 73856
block2_conv2 (Conv2D) (None, None, None, 128) 147584
block2_pool (MaxPooling2D) (None, None, None, 128) 0
block3_conv1 (Conv2D) (None, None, None, 256) 295168
block3_conv2 (Conv2D) (None, None, None, 256) 590080
block3_conv3 (Conv2D) (None, None, None, 256) 590080
block3_pool (MaxPooling2D) (None, None, None, 256) 0
block4_conv1 (Conv2D) (None, None, None, 512) 1180160
block4_conv2 (Conv2D) (None, None, None, 512) 2359808
block4_conv3 (Conv2D) (None, None, None, 512) 2359808
block4_pool (MaxPooling2D) (None, None, None, 512) 0
block5_conv1 (Conv2D) (None, None, None, 512) 2359808
block5_conv2 (Conv2D) (None, None, None, 512) 2359808
block5_conv3 (Conv2D) (None, None, None, 512) 2359808
=================================================================
Total params: 14,714,688
Trainable params: 14,714,688
Non-trainable params: 0
_________________________________________________________________
plt.imshow(im3_)
<matplotlib.image.AxesImage at 0x2c9da6fa0>
800/16
50.0
800/50 # 800 의 50 이면 16칸 거기에 중점은 8
16.0
x = np.arange(8, 800, 16) # 중점 8
y = np.arange(8, 800, 16)
x
array([ 8, 24, 40, 56, 72, 88, 104, 120, 136, 152, 168, 184, 200,
216, 232, 248, 264, 280, 296, 312, 328, 344, 360, 376, 392, 408,
424, 440, 456, 472, 488, 504, 520, 536, 552, 568, 584, 600, 616,
632, 648, 664, 680, 696, 712, 728, 744, 760, 776, 792])
center = np.array(np.meshgrid(x,y)).T.reshape(-1,2)
center
array([[ 8, 8],
[ 8, 24],
[ 8, 40],
...,
[792, 760],
[792, 776],
[792, 792]])
len(center)
2500
im4 = im3.copy()
for i in range(2500):
cv2.circle(im4, (center[i][0], center[i][1]), 1, (0,255,0), thickness=5)
plt.figure(figsize=(8,8))
plt.imshow(im4)
<matplotlib.image.AxesImage at 0x17e3c6400>
anchor_ratio = [0.5, 1, 2]
anchor_scale = [8, 16, 32]
anchor_list = np.zeros((2500*9,4), dtype=np.int32)
for i, c in enumerate(center):
c_x, c_y = c[0], c[1]
for r in anchor_ratio:
for s in anchor_scale:
h = pow(pow(s,2)/r, 0.5)
w = h*r
h *= 16
w *= 16
anchor_list[i] = [int(c_x-0.5*w), int(c_y-0.5*h), int(c_x+0.5*w), int(c_y+0.5*h)]
270*9
2430
anchor_list[2430]
array([ 413, 306, 1138, 669], dtype=int32)
im5 = im4.copy()
for i in range(2430, 2439):
cv2.rectangle(im5, (anchor_list[i][0],anchor_list[i][1]),
(anchor_list[i][2],anchor_list[i][3]), color=(0,0,255), thickness=3)
plt.figure(figsize=(8,8))
plt.imshow(im5)
<matplotlib.image.AxesImage at 0x2cc8d7310>
np.array(np.meshgrid(x,y)).reshape(-1,2)
array([[ 8, 24],
[ 40, 56],
[ 72, 88],
...,
[792, 792],
[792, 792],
[792, 792]])
# reference
a = np.arange(0,5)
b = np.arange(0,5)
np.meshgrid(a,b)[0]+np.meshgrid(a,b)[1]
array([[0, 1, 2, 3, 4],
[1, 2, 3, 4, 5],
[2, 3, 4, 5, 6],
[3, 4, 5, 6, 7],
[4, 5, 6, 7, 8]])
np.ogrid[0:5,0:5][0] + np.ogrid[0:5,0:5][1]
array([[0, 1, 2, 3, 4],
[1, 2, 3, 4, 5],
[2, 3, 4, 5, 6],
[3, 4, 5, 6, 7],
[4, 5, 6, 7, 8]])
np.mgrid[0:5,0:5]
array([[[0, 0, 0, 0, 0],
[1, 1, 1, 1, 1],
[2, 2, 2, 2, 2],
[3, 3, 3, 3, 3],
[4, 4, 4, 4, 4]],
[[0, 1, 2, 3, 4],
[0, 1, 2, 3, 4],
[0, 1, 2, 3, 4],
[0, 1, 2, 3, 4],
[0, 1, 2, 3, 4]]])
