Operating System – Water's Home

Cross compile wget with openssl for ARM

Posted on March 15, 2019March 15, 2019 by admin

Config OpenSSL

tar xvf openssl-1.1.1b.tar.gz 
cd openssl-1.1.1b/
mkdir __install
./config --prefix=$PWD/__install no-asm shared

Edit Makefile

PLATFORM=linux-armv4
CROSS_COMPILE=arm-linux-gnueabihf-
delete "-m32" "-m64"

Compile OpenSSL

make
make install

Config Wget && Make

tar zxvf wget-1.20.1.tar.gz
cd wget-1.20.1
./configure --prefix=$PWD/__install --host=arm-linux CC=arm-linux-gnueabihf-gcc --with-ssl=openssl --with-libssl-prefix=/root/Downloads/wget_arm/openssl-1.1.1b/__install --without-zlib
make 
make install

./xxx: /usr/lib/libstdc++.so.6: version `GLIBCXX_3.4.22′ not found (required by ./xxx)

Posted on March 15, 2019March 15, 2019 by admin

Copy libstdc++.so.6.0.24 from cross compiler (7.3.1)

cd /usr/lib
mv libstdc++.so.6 libstdc++.so.6.bak
ln -s libstdc++.so.6.0.24 libstdc++.so.6
chmod 777 libstdc++.so.6

Now, it will be like this !

root@imx6ul7d:/usr/lib# ls -al libstdc++.so.6*
lrwxrwxrwx 1 root root      19 Sep 16 23:03 libstdc++.so.6 -> libstdc++.so.6.0.24
-rwxr-xr-x 1 root root 1240644 Sep  9 02:04 libstdc++.so.6.0.21
-rwxrwxrwx 1 root root 9459568 Sep 16 23:35 libstdc++.so.6.0.24
lrwxrwxrwx 1 root root      19 Jan  1  1970 libstdc++.so.6.bak -> libstdc++.so.6.0.21
-rwxr-xr-x 1 root root 9459568 Sep 16 23:11 libstdc++.so.6.bak_24

TensorFlow on Android (Refer to codelabs : tensorflow-for-poets)

Posted on January 23, 2019January 23, 2019 by admin

The google demo is good, but there are some error when you follow it, so I copied and update it.

Install TensorFlow As Normal

```
pip install tensorflow==1.7.1
```

Download Google Demo

git clone https://github.com/googlecodelabs/tensorflow-for-poets-2

```
cd tensorflow-for-poets-2
```

Download TF_Files

wget http://download.tensorflow.org/example_images/flower_photos.tgz

```
tar -xvf flower_photos.tgz -C tf_files
```

Set ARCHITECTURE

```
IMAGE_SIZE=224
```

ARCHITECTURE="mobilenet_0.50_${IMAGE_SIZE}"

Start TensorBoard

tensorboard --logdir tf_files/training_summaries &

Help : retraining script

```
python -m scripts.retrain -h
```

Train

```
python -m scripts.retrain \
```

  --bottleneck_dir=tf_files/bottlenecks \

```
  --how_many_training_steps=500 \
```
```
  --model_dir=tf_files/models/ \
```

  --summaries_dir=tf_files/training_summaries/"${ARCHITECTURE}" \

  --output_graph=tf_files/retrained_graph.pb \

  --output_labels=tf_files/retrained_labels.txt \

```
  --architecture="${ARCHITECTURE}" \
```
```
  --image_dir=tf_files/flower_photos
```

Train with default 4,000 iterations

```
python -m scripts.retrain \
```

  --bottleneck_dir=tf_files/bottlenecks \

  --model_dir=tf_files/models/"${ARCHITECTURE}" \

  --summaries_dir=tf_files/training_summaries/"${ARCHITECTURE}" \

  --output_graph=tf_files/retrained_graph.pb \

  --output_labels=tf_files/retrained_labels.txt \

```
  --architecture="${ARCHITECTURE}" \
```
```
  --image_dir=tf_files/flower_photos
```

Train with Your Own Data

--image_dir= The root folder of the subdirectories which is used as label names by the classification script
This image is copied from google’s codelabs !

The Retrained Model

```
tf_files/retrained_graph.pb
```
```
tf_files/retrained_labels.txt
```

Test

(venv) water@water-G11CD:~/tensorflow-for-poets-2$ python -m scripts.label_image     --graph=tf_files/retrained_graph.pb      --image=tf_files/flower_photos/daisy/21652746_cc379e0eea_m.jpg

```
 
```

2019-01-23 14:13:50.234939: I tensorflow/core/platform/cpu_feature_guard.cc:140] Your CPU supports instructions that this TensorFlow binary was not compiled to use: AVX2 FMA

```
 
```
```
Evaluation time (1-image): 0.094s
```
```
 
```
```
daisy (score=0.99938)
```
```
sunflowers (score=0.00033)
```
```
dandelion (score=0.00028)
```
```
roses (score=0.00000)
```
```
tulips (score=0.00000)
```

Now, Prepare for Android

Install PILLOW

```
pip install PILLOW
```

Convert the model to TFLite format

```
IMAGE_SIZE=224
```

toco   --input_file=tf_files/retrained_graph.pb   --output_file=tf_files/optimized_graph.lite   --input_format=TENSORFLOW_GRAPHDEF   --output_format=TFLITE   --input_shape=1,${IMAGE_SIZE},${IMAGE_SIZE},3   --input_array=input   --output_array=final_result   --inference_type=FLOAT   --input_data_type=FLOAT

The TFLite Model

```
tf_files/optimized_graph.lite
```

Install AndroidStudio

unzip android-studio-ide-182.5199772-linux.zip

```
cd android-studio/
```
```
cd bin/
```
```
./studio.sh
```

Maybe You Need to Set Proxy

If you use lantern, you can do it like this!

Open Project

Open an existing Android Studio project

Choose android/tflite

Gradle Sync, YES

Build, Make Project

android/tflite/app/build/outputs/apk/debug/app-debug.apk

Test on Your Phone

Add your model files to the project, and then Make Project, Test Again

cp tf_files/optimized_graph.lite android/tflite/app/src/main/assets/graph.lite

cp tf_files/retrained_labels.txt android/tflite/app/src/main/assets/labels.txt

How to train YOLOv3 to detect Peppa Pig

Posted on January 22, 2019January 23, 2019 by admin

Prepare

Collecting Data from the Web with Python If you like

PASCAL VOC

.
├── VOCdevkit
│ └── VOC2018
│ ├── Annotations
├── ImageSets
│ │ └── Main // name of *.jpg without “.jpg”
├── JPEGImages // *.jpg

Annotation

LabelImg is a graphical image annotation tool

Example : XML

```
<annotation>
```
```
	<folder>000</folder>
```
```
	<filename>000000.jpg</filename>
```

	<path>/home/water/Machine_Learning/Peppa_Pig/train/000/000000.jpg</path>

```
	<source>
```
```
		<database>Unknown</database>
```
```
	</source>
```
```
	<size>
```
```
		<width>410</width>
```
```
		<height>256</height>
```
```
		<depth>3</depth>
```
```
	</size>
```
```
	<segmented>0</segmented>
```
```
	<object>
```
```
		<name>Peppa</name>
```
```
		<pose>Unspecified</pose>
```
```
		<truncated>0</truncated>
```
```
		<difficult>0</difficult>
```
```
		<bndbox>
```
```
			<xmin>64</xmin>
```
```
			<ymin>87</ymin>
```
```
			<xmax>166</xmax>
```
```
			<ymax>226</ymax>
```
```
		</bndbox>
```
```
	</object>
```
```
	<object>
```
```
		<name>Peppa</name>
```
```
		<pose>Unspecified</pose>
```
```
		<truncated>0</truncated>
```
```
		<difficult>0</difficult>
```
```
		<bndbox>
```
```
			<xmin>290</xmin>
```
```
			<ymin>77</ymin>
```
```
			<xmax>333</xmax>
```
```
			<ymax>131</ymax>
```
```
		</bndbox>
```
```
	</object>
```
```
</annotation>
```

Generate Labels for VOC

Darknet wants a .txt file for each image with a line for each ground truth object in the image that looks like:

```
<object-class> <x> <y> <width> <height>
```

voc_label.py

```
import xml.etree.ElementTree as ET
```
```
import pickle
```
```
import os
```
```
from os import listdir, getcwd
```
```
from os.path import join
```
```
 
```

sets=[('2018', 'train'), ('2018', 'val')]

```
 
```
```
classes = ["Peppa"]
```
```
 
```
```
def convert(size, box):
```
```
    dw = 1./size[0]
```
```
    dh = 1./size[1]
```
```
    x = (box[0] + box[1])/2.0
```
```
    y = (box[2] + box[3])/2.0
```
```
    w = box[1] - box[0]
```
```
    h = box[3] - box[2]
```
```
    x = x*dw
```
```
    w = w*dw
```
```
    y = y*dh
```
```
    h = h*dh
```
```
    return (x,y,w,h)
```
```
 
```
```
def convert_annotation(year, image_id):
```

    in_file = open('VOCdevkit/VOC%s/Annotations/%s.xml'%(year, image_id))

    out_file = open('VOCdevkit/VOC%s/labels/%s.txt'%(year, image_id), 'w')

```
    tree=ET.parse(in_file)
```
```
    root = tree.getroot()
```
```
    size = root.find('size')
```
```
    w = int(size.find('width').text)
```
```
    h = int(size.find('height').text)
```
```
 
```
```
    for obj in root.iter('object'):
```

        difficult = obj.find('difficult').text

```
        cls = obj.find('name').text
```

        if cls not in classes or int(difficult) == 1:

```
            continue
```
```
        cls_id = classes.index(cls)
```
```
        xmlbox = obj.find('bndbox')
```

        b = (float(xmlbox.find('xmin').text), float(xmlbox.find('xmax').text), float(xmlbox.find('ymin').text), float(xmlbox.find('ymax').text))

```
        bb = convert((w,h), b)
```

        out_file.write(str(cls_id) + " " + " ".join([str(a) for a in bb]) + '\n')

```
 
```
```
wd = getcwd()
```
```
 
```
```
for year, image_set in sets:
```

    if not os.path.exists('VOCdevkit/VOC%s/labels/'%(year)):

        os.makedirs('VOCdevkit/VOC%s/labels/'%(year))

    image_ids = open('VOCdevkit/VOC%s/ImageSets/Main/%s.txt'%(year, image_set)).read().strip().split()

    list_file = open('%s_%s.txt'%(year, image_set), 'w')

```
    for image_id in image_ids:
```

        list_file.write('%s/VOCdevkit/VOC%s/JPEGImages/%s.jpg\n'%(wd, year, image_id))

        convert_annotation(year, image_id)

```
    list_file.close()
```

peppa.data

```
classes= 1
```

train  = /home/d/Downloads/Peppa_VOC/2018_train.txt

valid  = /home/d/Downloads/Peppa_VOC/2018_val.txt

```
names = data/peppa.names
```
```
backup = backup
```

yolov3-voc-peppa.cfg

```
[net]
```
```
# Testing
```
```
batch=1
```
```
subdivisions=1
```
```
# Training
```
```
 batch=64
```
```
 subdivisions=2
```
```
width=416
```
```
height=416
```
```
channels=3
```
```
momentum=0.9
```
```
decay=0.0005
```
```
angle=0
```
```
saturation = 1.5
```
```
exposure = 1.5
```
```
hue=.1
```
```
 
```
```
learning_rate=0.001
```
```
burn_in=1000
```
```
max_batches = 500200
```
```
policy=steps
```
```
steps=400000,450000
```
```
scales=.1,.1
```
```
 
```
```
[convolutional]
```
```
batch_normalize=1
```
```
filters=16
```
```
size=3
```
```
stride=1
```
```
pad=1
```
```
activation=leaky
```
```
 
```
```
[maxpool]
```
```
size=2
```
```
stride=2
```
```
 
```
```
[convolutional]
```
```
batch_normalize=1
```
```
filters=32
```
```
size=3
```
```
stride=1
```
```
pad=1
```
```
activation=leaky
```
```
 
```
```
[maxpool]
```
```
size=2
```
```
stride=2
```
```
 
```
```
[convolutional]
```
```
batch_normalize=1
```
```
filters=64
```
```
size=3
```
```
stride=1
```
```
pad=1
```
```
activation=leaky
```
```
 
```
```
[maxpool]
```
```
size=2
```
```
stride=2
```
```
 
```
```
[convolutional]
```
```
batch_normalize=1
```
```
filters=128
```
```
size=3
```
```
stride=1
```
```
pad=1
```
```
activation=leaky
```
```
 
```
```
[maxpool]
```
```
size=2
```
```
stride=2
```
```
 
```
```
[convolutional]
```
```
batch_normalize=1
```
```
filters=256
```
```
size=3
```
```
stride=1
```
```
pad=1
```
```
activation=leaky
```
```
 
```
```
[maxpool]
```
```
size=2
```
```
stride=2
```
```
 
```
```
[convolutional]
```
```
batch_normalize=1
```
```
filters=512
```
```
size=3
```
```
stride=1
```
```
pad=1
```
```
activation=leaky
```
```
 
```
```
[maxpool]
```
```
size=2
```
```
stride=1
```
```
 
```
```
[convolutional]
```
```
batch_normalize=1
```
```
filters=1024
```
```
size=3
```
```
stride=1
```
```
pad=1
```
```
activation=leaky
```
```
 
```
```
###########
```
```
 
```
```
[convolutional]
```
```
batch_normalize=1
```
```
filters=256
```
```
size=1
```
```
stride=1
```
```
pad=1
```
```
activation=leaky
```
```
 
```
```
[convolutional]
```
```
batch_normalize=1
```
```
filters=512
```
```
size=3
```
```
stride=1
```
```
pad=1
```
```
activation=leaky
```
```
 
```
```
[convolutional]
```
```
size=1
```
```
stride=1
```
```
pad=1
```
```
filters=18
```
```
activation=linear
```
```
 
```
```
 
```
```
 
```
```
[yolo]
```
```
mask = 3,4,5
```

anchors = 10,14,  23,27,  37,58,  81,82,  135,169,  344,319

```
classes=1
```
```
num=6
```
```
jitter=.3
```
```
ignore_thresh = .7
```
```
truth_thresh = 1
```
```
random=1
```
```
 
```
```
[route]
```
```
layers = -4
```
```
 
```
```
[convolutional]
```
```
batch_normalize=1
```
```
filters=128
```
```
size=1
```
```
stride=1
```
```
pad=1
```
```
activation=leaky
```
```
 
```
```
[upsample]
```
```
stride=2
```
```
 
```
```
[route]
```
```
layers = -1, 8
```
```
 
```
```
[convolutional]
```
```
batch_normalize=1
```
```
filters=256
```
```
size=3
```
```
stride=1
```
```
pad=1
```
```
activation=leaky
```
```
 
```
```
[convolutional]
```
```
size=1
```
```
stride=1
```
```
pad=1
```
```
filters=18
```
```
activation=linear
```
```
 
```
```
[yolo]
```
```
mask = 0,1,2
```

anchors = 10,14,  23,27,  37,58,  81,82,  135,169,  344,319

```
classes=1
```
```
num=6
```
```
jitter=.3
```
```
ignore_thresh = .7
```
```
truth_thresh = 1
```
```
random=1
```

Train

./darknet detector train cfg/peppa.data cfg/yolov3-voc-peppa.cfg darknet53.conv.74

Test

Reference

```
@article{yolov3,
```

  title={YOLOv3: An Incremental Improvement},

  author={Redmon, Joseph and Farhadi, Ali},

```
  journal = {arXiv},
```
```
  year={2018}
```
```
}
```

Style Transfer : Caffe Example

Posted on January 17, 2019January 23, 2019 by admin

Default

Start

python nstyle.py

Example 1

Start with style-weight 2

python nstyle.py --content-image ~/Machine_Learning/Style_Transfer/src.png --style-image ~/Machine_Learning/Style_Transfer/style.jpg --output_dir ~/Machine_Learning/Style_Transfer/output_style-weight-2/ --save-epochs 200 --style-weight 2

Metric Learning : Caffe Example

Posted on January 17, 2019January 23, 2019 by admin

Prepare (MNIST Data)

```
python gen_pairwise_imglist.py
```

/home/water/caffe/build/tools/convert_imageset ./ train.txt train_lmdb --gray

/home/water/caffe/build/tools/convert_imageset ./ train_p.txt train_p_lmdb --gray

/home/water/caffe/build/tools/convert_imageset ./ val.txt val_lmdb --gray

/home/water/caffe/build/tools/convert_imageset ./ val_p.txt val_p_lmdb --gray

Based on Caffe siamese

Train loss vs iters

Visualize Result (Normal)

Visualize Result (τ- SNE)

Object Detection : MXNet SSD Example

Posted on January 17, 2019January 23, 2019 by admin

Prepare

```
cd /home/water/mxnet/example/ssd
```

wget https://github.com/zhreshold/mxnet-ssd/releases/download/v0.6/resnet50_ssd_512_voc0712_trainval.zi

unzip resnet50_ssd_512_voc0712_trainval.zip

python data/demo/download_demo_images.py

Test

```
python demo.py --gpu 0
```

python demo.py --epoch 0 --images ./data/demo/dog.jpg --thresh 0.5

python demo.py --cpu --network resnet50 --data-shape 512

```
. . . . . .
```

python demo.py --images ./data/demo/person.jpg --thresh 0.3

python demo.py --images ./data/demo/person.jpg --network inceptionv3

Result

Transfer learning & Fine tuning

Posted on January 15, 2019January 23, 2019 by admin

Based on CVGJ model

ResNet10_cvgj

Train Curve

Confusion Matrix

Precision-Recall, PR

Receiver operating characteristic curve, ROC

Visualize Activation

Collecting Data & Remove

Posted on January 15, 2019January 23, 2019 by admin

Collecting Data from the Web with Python If you like

Remove duplicate image

1. Remove invalid images

```
import os
```
```
import sys
```
```
import cv2
```
```
 
```

from collect_data import SUPPORTED_FORMATS

```
 
```
```
input_path = sys.argv[1]
```
```
 
```

for root, dirs, files in os.walk(input_path):

```
    for filename in files:
```

        ext = filename[filename.rfind('.')+1:].lower()

        if ext not in SUPPORTED_FORMATS:

```
            continue
```

        filepath = os.sep.join([root, filename])

        if cv2.imread(filepath) is None:

            os.system('rm {}'.format(filepath))

            print('{} is not a valid image file. Deleted!'.format(filepath))

2. Remove duplicate files

FDUPES is a program for identifying duplicate files residing
within specified directories.

```
sudo apt-get install fdupes
```
```
fdupes -rdN ./
```

3. Remove visually similar or duplicate images

3.1 Down scale

```
import os
```
```
import cv2
```
```
import sys
```
```
 
```
```
input_path = sys.argv[1].rstrip(os.sep)
```
```
 
```
```
target_short_edge = int(sys.argv[2])
```
```
 
```

for root, dirs, files in os.walk(input_path):

    print('scanning {} ...'.format(root))

```
    for filename in files:
```

        filepath = os.sep.join([root, filename])

```
        img = cv2.imread(filepath)
```
```
        h, w = img.shape[:2]
```
```
        short_edge = min(w, h)
```
```
 
```

        if short_edge > target_short_edge:

            scale = float(target_short_edge) / float(short_edge)

            new_w = int(round(w * scale))

            new_h = int(round(h * scale))

            print('Down sampling {} from {} x {} to {} x {} ...'.format(

                filepath, w, w, new_w, new_h

```
            ))
```

            img = cv2.resize(img, (new_w, new_h))

```
            cv2.imwrite(filepath, img)
```
```
print('Done!')
```

3.2 Find image dupes

```
sudo apt-get install findimagedupes
```
```
findimagedupes -R train > dup_list
```

3.3 Remove duplicates

```
import os
```
```
import sys
```
```
dup_list = sys.argv[1]
```
```
with open(dup_list, 'r') as f:
```
```
    lines = f.readlines()
```
```
    for line in lines:
```
```
        dups = line.split()
```

        print('Removing duplicates of {}'.format(dups[0]))

```
        for dup in dups[1:]:
```
```
            cmd = 'rm {}'.format(dup)
```
```
            os.system(cmd)
```

Regression : Caffe

Posted on January 15, 2019January 23, 2019 by admin

Based on LeNet-5

Deploy with GPU

```
name: "RegressionExample"
```
```
layer {
```
```
  name: "data"
```
```
  type: "Input"
```
```
  top: "data"
```
```
  input_param    {
```
```
    shape:  {
```
```
        dim:    100
```
```
        dim:    1
```
```
        dim:    100
```
```
        dim:    100
```
```
    }
```
```
  }
```
```
}
```
```
……
```

Config OpenSSL

Edit Makefile

Compile OpenSSL

Config Wget && Make

Copy libstdc++.so.6.0.24 from cross compiler (7.3.1)

Now, it will be like this !

Install TensorFlow As Normal

Download Google Demo

Download TF_Files

Set ARCHITECTURE

Start TensorBoard

Help : retraining script

Train

Train with default 4,000 iterations

Train with Your Own Data

The Retrained Model

Test

Now, Prepare for Android

Install PILLOW

Convert the model to TFLite format

The TFLite Model

Install AndroidStudio

Maybe You Need to Set Proxy

Open Project

Open an existing Android Studio project

Choose android/tflite

Gradle Sync, YES

Build, Make Project

Test on Your Phone

Add your model files to the project, and then Make Project, Test Again

Prepare

PASCAL VOC

Annotation

Example : XML

Generate Labels for VOC

voc_label.py

peppa.data

yolov3-voc-peppa.cfg

Train

Test

Reference

Default

Start

Example 1

Start with style-weight 2

Prepare (MNIST Data)

Based on Caffe siamese

Train loss vs iters

Visualize Result (Normal)

Visualize Result (τ- SNE)

Prepare

Test

Result

Based on CVGJ model

Train Curve

Confusion Matrix

Precision-Recall, PR

Receiver operating characteristic curve, ROC

Visualize Activation

Remove duplicate image

1. Remove invalid images

2. Remove duplicate files

3. Remove visually similar or duplicate images

3.1 Down scale

3.2 Find image dupes

3.3 Remove duplicates

Based on LeNet-5

Deploy with GPU

Convolution Visualization