Trjtdtk

Is CEO the "profession" with the most psychopaths?

How does the math work when buying airline miles?

Is there a kind of relay only consumes power when switching?

Redirect to div in page with #

Sum letters are not two different

Why is it faster to reheat something than it is to cook it?

Would "destroying" Wurmcoil Engine prevent its tokens from being created?

Has negative voting ever been officially implemented in elections, or seriously proposed, or even studied?

Disembodied hand growing fangs

Is there any word for a place full of confusion?

How do I make this wiring inside cabinet safer?

AppleTVs create a chatty alternate WiFi network

QGIS: how to apply Line Pattern Fill to LineStrings?

Why wasn't DOSKEY integrated with COMMAND.COM?

Project Euler #1 in C++

As a beginner, should I get a Squier Strat with a SSS config or a HSS?

If windows 7 doesn't support WSL, then what does Linux subsystem option mean?

Circuit to "zoom in" on mV fluctuations of a DC signal?

Is it ethical to give a final exam after the professor has quit before teaching the remaining chapters of the course?

Did MS DOS itself ever use blinking text?

How often does castling occur in grandmaster games?

What do you call the main part of a joke?

How to convince students of the implication truth values?

Run Powershell Script From Agent As Administrator

Bounding box regression in R-CNN

0

$begingroup$

In R-CNN paper, they give the definition of the target values for bounding box regression

Given that $(P, G)$ is a (prediction box, ground-truth box) pair of the form $(x, y, w, h)$ where $x, y$ is the center coordinate of the box, $w, h$ are width and height respectively.

$t_x = (G_x - P_x) / P_w hspace2.0cm t_y = (G_y - P_y) / P_h$

$t_w = log(G_w / P_w) hspace2.0cm t_h = log(G_h / P_h)$

And the goal is to find $textbfw_*$, where $*$ can be $x, y, w$ or $h$, so that

$textbfw_* = arg min_hattextbfw_* sum_i (t^i_* - hattextbfw_*^T phi(P^i))^2 + lambda |hattextbfw_*|^2$ where $phi(P^i)$ is the feature map given by the last pooling layer of the feature extractor after passing predicted bounding box $P^i$

I don't understand why they come up with this approach of bounding box regression ? Can anyone tell me about this ?

P.S: since this regression approach is used not only in R-CNN but also in later models, I really want to get a clear understanding of this

cnn object-detection faster-rcnn

share|improve this question

asked Apr 3 at 1:15

HOANG GIANGHOANG GIANG

426

$endgroup$

add a comment | 

0

$begingroup$

In R-CNN paper, they give the definition of the target values for bounding box regression

Given that $(P, G)$ is a (prediction box, ground-truth box) pair of the form $(x, y, w, h)$ where $x, y$ is the center coordinate of the box, $w, h$ are width and height respectively.

$t_x = (G_x - P_x) / P_w hspace2.0cm t_y = (G_y - P_y) / P_h$

$t_w = log(G_w / P_w) hspace2.0cm t_h = log(G_h / P_h)$

And the goal is to find $textbfw_*$, where $*$ can be $x, y, w$ or $h$, so that

$textbfw_* = arg min_hattextbfw_* sum_i (t^i_* - hattextbfw_*^T phi(P^i))^2 + lambda |hattextbfw_*|^2$ where $phi(P^i)$ is the feature map given by the last pooling layer of the feature extractor after passing predicted bounding box $P^i$

I don't understand why they come up with this approach of bounding box regression ? Can anyone tell me about this ?

P.S: since this regression approach is used not only in R-CNN but also in later models, I really want to get a clear understanding of this

cnn object-detection faster-rcnn

share|improve this question

asked Apr 3 at 1:15

HOANG GIANGHOANG GIANG

426

$endgroup$

add a comment | 

$begingroup$

In R-CNN paper, they give the definition of the target values for bounding box regression

Given that $(P, G)$ is a (prediction box, ground-truth box) pair of the form $(x, y, w, h)$ where $x, y$ is the center coordinate of the box, $w, h$ are width and height respectively.

$t_x = (G_x - P_x) / P_w hspace2.0cm t_y = (G_y - P_y) / P_h$

$t_w = log(G_w / P_w) hspace2.0cm t_h = log(G_h / P_h)$

And the goal is to find $textbfw_*$, where $*$ can be $x, y, w$ or $h$, so that

$textbfw_* = arg min_hattextbfw_* sum_i (t^i_* - hattextbfw_*^T phi(P^i))^2 + lambda |hattextbfw_*|^2$ where $phi(P^i)$ is the feature map given by the last pooling layer of the feature extractor after passing predicted bounding box $P^i$

I don't understand why they come up with this approach of bounding box regression ? Can anyone tell me about this ?

P.S: since this regression approach is used not only in R-CNN but also in later models, I really want to get a clear understanding of this

cnn object-detection faster-rcnn

share|improve this question

asked Apr 3 at 1:15

HOANG GIANGHOANG GIANG

426

$endgroup$

share|improve this question

asked Apr 3 at 1:15

HOANG GIANGHOANG GIANG

426

share|improve this question

asked Apr 3 at 1:15

HOANG GIANGHOANG GIANG

426

asked Apr 3 at 1:15

HOANG GIANGHOANG GIANG

426

asked Apr 3 at 1:15

HOANG GIANGHOANG GIANG

426