技术

Deep learning memo

by

yolo pdf,

EfficientDet	https://arxiv.org/pdf/2011.08036.pdf
YOLOv1		https://arxiv.org/pdf/1506.02640.pdf
YOLOv2		https://arxiv.org/pdf/1612.08242.pdf
YOLOv3		https://arxiv.org/pdf/2011.08036.pdf
		https://pjreddie.com/media/files/papers/YOLOv3.pdf
YOLOv4		https://github.com/AlexeyAB/darknet
YOLOv4-Scaled	https://github.com/WongKinYiu/ScaledYOLOv4
YOLO-PPv2	https://arxiv.org/pdf/2104.10419.pdf
YOLOv5		https://arxiv.org/pdf/2104.10419.pdf
YOLOX		https://github.com/Megvii-BaseDetection/YOLOX
YOLOR		https://github.com/WongKinYiu/yolor
YOLOF		https://arxiv.org/pdf/2103.09460.pdf
YOLOS		https://arxiv.org/pdf/2106.00666.pdf
YOLOP		https://arxiv.org/pdf/2108.11250.pdf
YOLOV6		https://mp.weixin.qq.com/s/RrQCP4pTSwpTmSgvly9evg
		https://github.com/meituan/YOLOv6 
		https://zhuanlan.zhihu.com/p/353697121 
 
YOLOV7		https://arxiv.org/abs/2207.02696 
		https://github.com/WongKinYiu/yolov7

https://github.com/yael-vinker/CLIPasso
https://github.com/openai/CLIP

refer to:
AI有啥用
https://blog.csdn.net/PercentageC/article/details/126744534

Speed up pip install

by 
1
2
3
4
5
6
7
8
9

python3 -m pip install opencv-python==4.5.3.56 -i https://pypi.tuna.tsinghua.edu.cn/simple
 
pip config set global.index-url https://pypi.tuna.tsinghua.edu.cn/simple
 
http://mirrors.aliyun.com/pypi/simple/ 
https://pypi.mirrors.ustc.edu.cn/simple/ 
https://pypi.douban.com/simple/ 
https://pypi.tuna.tsinghua.edu.cn/simple/ 
http://pypi.mirrors.ustc.edu.cn/simple/

refer to:
https://www.cnblogs.com/huangguifeng/p/12002049.html

Chrome memo

by

To prevent inactive tab from using cpu resources, in address bar, visit

1

chrome://discards

toggle restless pages 'Urgent Discard' to grey under 'Actions' column.

chrome://flags#

enable-parallel-downloading
enable-force-dark

refer to:
https://www.askvg.com/tip-enable-tab-freeze-or-tab-suspend-feature-in-google-chrome/
https://blog.csdn.net/ythuiyi/article/details/113506238

Rendering mp4 video not silky, why?

by

The original HD mp4 file 'movie.mp4' plays very smoothly. I picked out the h264 stream saved as 'movie.264' from it, then using mp4v2 converted 'movie.264' to a new mp4 file 'output.mp4'. However, experience of watching 'output.mp4' was very bad. Why?

I debugged the 'ClipTrack' function in 'mp4v2/test/OLD/mp4clip.cpp', found a key argument 'renderingOffset' of function 'MP4WriteSample'. All examples searched from the web about writing h264 to mp4 through mp4v2 set 'renderingOffset' to 0, while in 'ClipTrack', 'renderingOffset' of the function 'MP4WriteSample' is directly set to the value from 'renderingOffset' of the function 'MP4ReadSample'.

Here we make a concept clear, the 'sample' entity in mp4 file format specification is equal to a frame which can completely rendering as a picture. A sample consists of several nalus, a nalu whose start code is 0x00, 0x00, 0x00, 0x01 indicates a new sample, while a nalu which starts with 0x00, 0x00, 0x01 is among a sample, and a sample needs not to be a IDR key frame.

Through carrying 'renderingOffset' value at beginning of each sample in 'movie.264', I did generate an 'output.mp4' which displays as smoothly as the original one.

Did 'renderingOffset' mean time stamp of a sample? No, at least not exactly. By watching log, I found its value was not increasing as time went by.

I continued debugging mp4v2, then saw a key word 'ctts' atom, and searched a link from the web,

https://blog.csdn.net/cheyo809775692/article/details/100924523

I now understand, the decoding order of h264 samples is slightly different from the rendering order of them.

For example, in stts atom,

1
2
3
4
5

'decoding order'	'sample count'		'sample duration'
1,2,3,4			4			20
5			1			40
6,7,8			3			20
9			1			60

in ctts atom,

1
2
3
4
5
6
7
8

'decoding order'	'sample count'		'sample offset'
1,2,3			3			20
4			1			40
5			1			0
6			1			40
7			1			0
8			1			40
9			1			0

final rendering timing is

1
2
3

'decoding order'	1	2	3	4	5	6	7	8	9
'duration'		20	20	20	20	40	20	20	20	60
'rending order'			1	2	3	5	5,4	7	6	9	9,8	9

Terms,

1
2
3
4
5
6
7
8
9
10
11
12

mvhd: movie header
tkhd: track header
mdhd: media header
stsd: sample descriptions
stts: time to sample
ctts: composition offset
stss: sync sample offset
stsc: sample to chunk
stsz: sample size
stco/stco64: chunk offset
dts: decoding timestamp
pts: presentation timestamp

expect scp

by 
1
2
3
4
5
6
7
8
9
10
11
12
13
14

auto_scp()
{
	local filepath=$1
	expect -c "
		set timeout 10
 
		spawn scp scp://someone@192.168.12.3:1234//home/someone/$filepath .
 
		expect \"password:\" 
		send \"toodangerous\r\"
 
		interact
	"
}

refer to:
https://blog.csdn.net/f_zyj/article/details/93475830
https://zhuanlan.zhihu.com/p/25164676

Select Windows SDK version when building using cmake

by

Read cmake online doc carefully, https://cmake.org/cmake/help/latest/variable/CMAKE_VS_WINDOWS_TARGET_PLATFORM_VERSION.html

So when we build project which has CMakeLists.txt, we can select Windows SDK version in this way:

  1. In explorer, navigate to like C:\Program Files (x86)\Windows Kits\10\Include.
  2. For cmake always finds the newest version to build our project, we can just rename all sdk versions which is newer than we want. e.g. if we want 10.0.19041.0 to build our project, but 10.0.22581.0 is newer, we rename 10.0.22581.0 to 10.0.0.22581.0, the rule is among tokens split by period signs, the third token number is the key word that cmake sorts them to get the largest as newest.
  3. Start cmake to configure the project, it will find the right sdk version.
  4. After cmake is done in configuration, rename 10.0.0.22581.0 back to 10.0.22581.0.

LD_LIBRARY_PATH in makefile

by

At a glance,

1
2
3

dbg:
	set LD_LIBRARY_PATH=/path/to/so
	echo ${LD_LIBRARY_PATH}

Why does the 'echo' command output nothing?

For the dollar sign is not only a key letter in bash, it's also in makefile. And each command line in Makefile shares no environmental variables. So right way is

1
2
3

dbg:
	set LD_LIBRARY_PATH=/path/to/so ; \
	echo $${LD_LIBRARY_PATH}

However, if tst01 needs to load libtst02.so from /path/to/so/,

1
2

dbg:
	set LD_LIBRARY_PATH=/path/to/so ; ./tst01

why does tst01 still report it can't find libtst02.so?

For the 'set' command only lets LD_LIBRARY_PATH variable visible in the bash process where the 'set' is invoked, no spreading in child processes. So final correct script is

1
2

dbg:
	export LD_LIBRARY_PATH=/path/to/so ; ./tst01

logc needs rewritting

by

For the thing is changing, loge will not call logi by itself, we need invoke !logexts.logi explicitly in advance.

The up to date procedure is

  1. LogManifest.lgm is in wrong place, so

    1
2

    cd C:\Program Files (x86)\Windows Kits\10\Debuggers\x64\winext
copy LogManifest.lgm ..
  2. Start WinDbg, click main menu, File, Open Executable...
  3. In command bar of WinDbg, input

    1
2
3
4

    !logexts.logi
g
!logir
g

We will find a folder named ApiLogs lying on the Windows desktop. After the debuggee process exits, by using logviewer.exe we can open files in it, with .lgx suffix.

Automatically distill headers from other project

by

catch_inc.sh

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44

#!/bin/bash
 
dir_cur=`pwd`
dir_inc=original_inc
dir_src=/path/to/other/project
 
withly()
{
	[ -e $2 ] && $*
}
 
withoutly()
{
	[ -e $2 ] || $*
}
 
withly rm inc.makefile1
withly rm Makefile
 
if [ x$1 == xclean ] ; then
	withly rm inc.makefile
	withly rm ${dir_inc} -Rf
	exit 0
fi
 
withoutly mkdir ${dir_inc}
withoutly touch inc.makefile
 
while [ 1 ] ; do
	cat inc.makefile | sort | uniq > inc.makefile1
	cp inc.makefile1 inc.makefile
	cat inc.makefile | sed ':t;N;s/\n//;b t' | sed 's/\//\\\//g' | sed 's/\ /\\\ /g' > inc.makefile1
	sed "s/@original_includes@/`cat inc.makefile1`/g" Makefile.pat > Makefile
	lost_header=`make 2>&1 | grep "#include" | awk '{if ($2 == "|" && $3 == "#include") print $4}' | awk -F\" '{print $2}'`
	if [ x${lost_header} == x ] ; then
		break
	fi
	found_header=`cd ${dir_src}; find . -name ${lost_header} | head -n 1`
	echo ${found_header}
	found_header=${found_header##./}
	( cd ${dir_src}; cp --path ${found_header} ${dir_cur}/${dir_inc}/ )
	found_header=`dirname ${found_header}`
	echo "-I${dir_inc}/${found_header} " | cat >> inc.makefile
done

For an example, pattern file Makefile.pat is as

1
2
3
4

...
CFLAGS=-g -Iinclude @original_includes@
LDFLAGS=-pthread -lm
...

from which the catch_inc.sh script can generate Makefile like

1
2
3
4

...
CFLAGS=-g -Iinclude -Ioriginal_inc/subdir/in/other/project/inc1 -Ioriginal_inc/subdir/in/other/project/common/inc2
LDFLAGS=-pthread -lm
...

refer to:
https://www.cnblogs.com/liqiu/p/4506508.html

Hardware memo

by 
1
2
3
4
5
6
7
8

PWM	Pulse Width Modulation
SPI	Serial Peripheral Interface			MISO MOSI SCLK CS
UART	Universal Asynchronous Receiver/Transmitter	Rx Tx GND
USB	Universal Serial Bus				VBUS D- D+ GND
AMBA	Advanced Microcontroller Bus Architecture	AHB ASB APB
MIPI	Mobile Industry Processor Interface
LVDS	Low Voltage Differential Signaling
CFA	Color Filter Arrays

HiSi

1
2
3
4
5
6
7
8
9

NNIE	DPU
IVE	ISP
VI	Video Input
VO	Video Output
VPSS	Video Processing Sub System
AVS	Video Splicing
MMZ	Media Memory Zone
VB	Video Buffer
NR	Noise Reduction

refer to:
https://www.cnblogs.com/qinghaowusu/p/13610568.html