004_comission/vinniesniper-54816/task1/_lab/journal.md

vnc

host 192.168.10.21

http://192.168.10.21:6199/

work path /home/logic/_workspace/task-list/servers/logic-NUC8i5BEH/opencv-workdesk/001_monitor/src/007-test

test results

currently bad result on recognizing elephant, horse and mountain

• i want to know if increase nPos can help

  • 400 -> 4000
  • 400 -> 12000

• i am testing nPos 10, nNeg ~2000, nStages 15 on 007-train-bus-test-big-pos-pic

  • i want to know if large dimensions of pos image can help

nPos(p)	nNeg(n)	nStage(s)	px	success	TT	TT base	FT	FT base	reasons / remarks / conclusion
12800	200	5	24	failed	?	?	?	?	007-train-bus-test-pos-pic-100px, i want to know lower stage can build ?
32000	200	5	24	failed	?	?	?	?	007-train-bus-test-pos-pic-24px, i need a build success
9600	200	5	24	failed	?	?	?	?	007-train-bus-test-pos-pic-100px, i want to know lower stage can build ?
9600	200	15	24	failed	?	?	?	?	007-train-bus-test-pos-pic-24px, i need a build success
2400	200	15	24	failed	?	?	?	?	007-train-bus-test-pos-pic-24px, i need a build success
400	200	15	50	N/A	1	50	2	450	007-train-bus-test-big-pos-pic, seems good when using 50px, build failed, increase nPos to 800
50	200	15	100	N/A	50	50	450	450	007-train-bus-test-pos-pic-50, try pos 10 -> 50 samples,
10	200	5	100	N/A	0	50	0	450	007-train-bus-test-pos-pic-100px, try 100px pos size
10	200	15	50	N/A	1	50	2	450	007-train-bus-test-big-pos-pic, try 50px pos size
10	200	5	orig. size	N/A	0	50	0	450	seems original size not a correct size, try 50px black and white
200	1783	15	24	N/A	7	7	51	51	ongoing, test to find bus image
200	1971	15	24	N/A	7	7	51	51	ongoing, add flowers and horse
200	1603	5	50	N/A	x	x	x	x	try reduce fault rate by increase px, resut -> cannot built
200	1603	10	24	N/A	50	50	441	450	stage 10 is not acceptable, return to use 15
200	1603	5	24	N/A	50	50	450	~450	small stage can kills accurancy
200	1603	15	24	N/A	33	50	100	450	currently best bulid
200	1354	5	24	N/A	4	4	9	26	i want to see if stage can keep same result but trainging faster
200	1354	15	24	N/A	4	4	9	26	ongoing
200	944	15	24	N/A	4	4	9	26	guess, this is 900 + 44 bus photo result
200	900	15	100	N/A	2	2	7	14	ongoing, i want to know if i increase px increase accuracy
200	900	15	24	N/A	2	2	7	14
100	300	15	x	N/A	33	142	x	x

-numPos 100
-numNeg 300
-numStages 15

count_good_result      : 33
count_good_result_base : 99
count_bad_result       : 142

test with 100 posdata

count_good_result      : 33
count_good_result_base : 99
count_bad_result       : 142

https://pythonprogramming.net/haar-cascade-object-detection-python-opencv-tutorial/

i want to test haar cascade classifier, i need to make a helloworld

https://machinelearningmastery.com/training-a-haar-cascade-object-detector-in-opencv/

https://github.com/yang-hfff/opencv_traincascade/tree/master

https://cloud.tencent.com/developer/article/1996918