【ML Paper】Explanation of all of YOLO series Part 6

This is an summary of paper that explains yolov1 to v8 in one.
Let's see the history of the yolo with this paper.
This article is part 6, part 5 is here.

Original Paper: https://arxiv.org/pdf/2304.00501

4.2 YOLOv1 Architecture

Architecture Overview

The YOLOv1 architecture consists of 24 convolutional layers followed by two fully-connected layers that predict bounding box coordinates and probabilities. Inspired by GoogLeNet and Network in Network, YOLO incorporates $1 \times 1$ convolutional layers to reduce the number of feature maps, thereby maintaining a relatively low number of parameters.

Activation Functions

All layers in YOLOv1 utilize leaky rectified linear unit (Leaky ReLU) activations, except for the final layer, which employs a linear activation function. The activation layers are detailed in Table 1.

Model Variants

The authors also introduced a lighter version of the model called Fast YOLO, which is composed of nine convolutional layers. This variant offers a more streamlined architecture suitable for scenarios requiring reduced computational complexity.