The purpose of this post is to share my impressions of writing a project, from the implementation of which I got a share of adrenaline, and to help DLS students in their projects who have chosen the topic of detection. And in general - to share ideas.
Small introduction
With the outbreak of the pandemic, my life changed dramatically. Like many others, I was sent to a remote location. At the same time, investments in the project decreased sharply and I had more free time. I began to improve my self-education, took two or three courses in steppe and there I noticed the course of the Deep Learning School, or DLS for short. In the course of the training, I passed the NLP course from Huawei in parallel, and began the course at the School-IT from BEAC. In general, I decided to slightly shift the vector of activity towards AI. And now I got to the final DLS project. There's also an Alpha Wars competition on the way . In general, that Autumn-Winter Marathon. But - that's a topic for another article. So, you need to force events in order to do everything.
The first day
A few words about myself: I have good experience with PHP backend development, but my last project was implemented in Python. I also had to implement different projects or parts of them in C / C ++, lua and even js. The last one, I dislike much since IE 3.0, when there was no debugging and the code kept crashing. Since then, by hook or by hook I have been trying to get away from the front-end. But, now the time requirements are such that without the use of AJAX, there is simply no place, so you have to know JQuery to a minimum.
I digress a little, but to make it more clear, when the topics of graduation projects were announced, the choice definitely fell on the WEB project. I knew in advance how to make the WEB part, similar patterns were implemented more than once, and even spoke on this topic at the Conference
" ".
:
WEB
, ...
, detection , 100500+ , - , 50 . . : nvideo_ssd mmdetection. . .
, !! - . , , , WEB ... , , : GPU, CPU. CPU , - . , - . , .
mmdetection. , GPU. - CPU. !
mmdetection . . readme GPU, CPU . , -... , - mmdetection.py, .
WEB . hhtp server , , - , pyramid. WEB- . , - index.html .
index.html :
AJAX
WEB , png , . . : sid = 123, 123.png
, Js AJAX [sid].out.png.
[sid].out.png , Js .
! - . WEB , [sid].out.png . WEB .
. WEB . : POST . , .
, WEB .
? , CPU 20-30 , . . WEB , WEB (WEB ) , WEB .
, , . , . , , - . , mmdetection miniconda, miniconda . miniconda - .
, : , . . , , . . , . Z- . - processlist. , , , processlist . - Z-f - .
? , , sid, , - , . :
.
AJAX
WEB , png , . . : sid = 123, 123.png
WEB (Js ) sid, 7 .
WEB : mmdetectionl.py .
sid .
mmdetectionl.py : [sid].out.png: 123.out.png.
, Js AJAX [sid].out.png.
[sid].out.png , Js .
.
- ?
. , MVP. , , - . , , - . . , , . , , .
. JQuery, . . HTTP .
Don't judge the js code, I know what shit is. With python code a little better, although not in PES format
Link to github - removed before the end of the deadline
Link ha project Design - a word with the letter G, no time and no design task ...
The project may not work. There may be two reasons: either I am improving something in it, then contact telegram @akalend Second: this is a test server from work, and it can be taken away at any time for other needs or I am fired. Any comments are appreciated. Hope this helps someone.
PS
for those who make bot telegrams - the principle is the same.