Save images

[tinhnguyen-starglobal]

• photos in png format
• output at 2X resolution

[tinhnguyen-starglobal]

LGTM!

[tinhnguyen-starglobal]

Hi @fritz-ebner please help me review this PR. Thanks

[graySquirrel]

I don't think that is correct, but you will have to check with Dmytro.
I think the code is line 48, should not do sigmoid(tc)

[tinhnguyen-starglobal]

Hi @fritz-ebner as you can see,
let confidence = sigmoid(tc)
-> we create a constant "confidence" here and assigned sigmoid(tc) to it.
But I found that is reason, that led to a low confidence score. So I ignored the "confidence" value and didn't use it anymore.

To avoid getting you confused I will remove this line of code. Please help me check.

[fritz-ebner]

I think line 57 still has to be
let confidenceInClass = tc * bestClassScore

[tinhnguyen-starglobal]

Hi @fritz-ebner
In the beginning, I intend to use: let confidenceInClass = tc * bestClassScore
But Alexey and Dmytro Doroshenko explained we should use: let confidenceInClass = tc
Because: tc values in range [0,1] and they are bounding box score, while bestClassScore is a probability of a specific class on this (current) bounding box.
If you need more detail please let me know. thanks.