All language subtitles for 002 YOLOv7 architecture detailed explanation

In this video, I will explain about the architecture of your office seven.

Here is the architecture of YOLO v seven.

In the backbone computational blocks YOLO v seven using e lan and e lan, but e line is only used on

all of 76 c in the net of e seven sensors p p to sp exp.

YOLO v seven also employs an optimize path aggregation network by incorporating e lan.

On the head.

YOLO V seven integrates three scales based on the neck with additional rib conf.

The big ones.

Architectural details are as follows CB's is mainly composed of convolution bells, normalization and

similar activation function.

KBS connects better normalization layer directly to convolutional layer.

The purpose of this is to integrate the mean and variance of bats normalization into the bias and weight

of convolutional layer two Inference states.

YOLO v seven As previously stated implies, you learn in its computational blocks.

Here are the details from Ilan.

Next there is empty conf.

The NP conflict is mainly divided into Max Pro and CBSE.

Next on the neck.

First, I will explain what spe pxp.

Here are the details from SP PXP.

The SP SP module adds the correct operation at the end based on the SP module, which is fused with

the FITS before the SP module.

It aims to enrich the feature information.

The following are the details of the SP.

YOLO v seven also uses an optimised path aggregation network that incorporates inline.

Path aggregation network and student sample operation after sample.

Power Aggregation network was chosen because of its ability to accurately preserve spatial information

which aids in the proper localization of pixels.

Next, here are the details of the event on the NEC.

On the head of seven integrates three scales based on the neck and allocates three anchor boxes under

each scale.

By using three scales, it increases accuracy when detecting three object sizes small, medium and large.

In addition, we call this also added.

Web.com refused to change the number of channels.

Output from webcomic has a certain difference between training and inference.

There is an additive output of the three branches during training and the parameters of the branches

are parameterized to the main brands during deployment.

That's the explanation of the YOLO v seven architecture.

See you in the next video.