training a model about mobilenet-ssd HOT 8 CLOSED

@3073
Let's organize one by one.

but it skip the detection of the object and even sometimes it does not detect at all.

First of all, please adjust the "threshould" number in program.

For example)
threshould = 0.80 → 0.45 → 0.30 → 0.25 → ...
thresh = 0.7 → 0.3 .→ ...
min_score_percent = 0.8 → 0.7 → 0.6 → 0.5 → ...

"Threshold(min_score_percent )" means the threshold of probability of being "human".

from mobilenet-ssd.

from mobilenet-ssd.

@3073
Please change to 0.2

from mobilenet-ssd.

from mobilenet-ssd.

@3073
I think you noticed.
If threshould is lowered, extra objects are detected, raising threshould will decrease the detection rate.
You need to decide on reasonable "threshould" yourself.
If you are not satisfied with any "threshould", It is necessary to re-train the model or review the model itself.
First of all, please try looking for yourself whether there is no satisfying "threshould".
The important thing is not to ask for perfection.

from mobilenet-ssd.

from mobilenet-ssd.

@3073
I do not mean to confuse you.
I got your time a little to find out what you understand.

OK.
For details on how to train model yourself, please see below.
It is written in my "Readme" from the beginning, so please read it carefully.

https://github.com/PINTO0309/MobileNet-SSD-RealSense#construction-of-learning-environment-and-simple-test-for-model-ubuntu1604-x86_64-pc--gpunvidia-geforce

or

https://github.com/avBuffer/MobilenetSSD_caffe

from mobilenet-ssd.

@3073
One month passed since the last answer.
Since there seems to be no particular problem, I will close "issue" once.

from mobilenet-ssd.