README

---

Abstract：这个部分用于使用传统图像特征的方法，基于滑动窗口的的目标检测 模块，包含Config模块，FeatureExtractor模块，和Classifier模块，下面将简单 介绍使用方法和各模块的功能。

Acknowledge

• Code framework from Zhongyang Zhang

使用方法

训练

python main.py --action train

测试

python main.py --action test

各模块功能

Config	FeatureExtractor	Classifier
定义project的各种参量，包括使用的feature类型，project_id，路径（训练集/
测试集/模型）以及各种feature提取的参数	定义特征提取函数，该模块包含通用的
图像处理函数，以及特定的feature提取函数	定义分类器的训练以及预测函数

Config

Param	Definition
DES_TYPE	The feature extraction type
CLF_TYPE	The classifier model(eg.SVM,MLP)
project_id	Self_defined project name
THRESHOLD	The threshold when applying nms()
DOWNSCALE	The downscale when applying pyramid_gaussian()
MIN_WDW_SIZE	The min size of windows detected
STEP_SIZE	The slide step when sliding across the image

FeatureExtractor

Function	Definition
resize_crop_by_short()	Resize and crop the input image, output image shape(short_len, short_len)
resize_by_short()	Resize the image
image_preprocess_size()	Resize and crop all training images
sliding_window()	Slide window at a fix window size
overlapping_area()	Calculate the overlap area of two detections
nms()	Apply NMS
process_image()	Extract features from input image
extract_features()	Extract features of all input images

Classifier

Function	Definition
load_data()	Load all features of images
train_classifier()	Train the model
load_model()	Load the model
predict()	Predict the class or score of the input image
test_classifier()	Test classifier on the test images