这是一个yolox-pytorch的源码，可以用于训练自己的模型。

Related tags

Deep Learningyolox-pytorch
Overview

YOLOX：You Only Look Once目标检测模型在Pytorch当中的实现

目录

  1. 性能情况 Performance
  2. 实现的内容 Achievement
  3. 所需环境 Environment
  4. 小技巧的设置 TricksSet
  5. 文件下载 Download
  6. 训练步骤 How2train
  7. 预测步骤 How2predict
  8. 评估步骤 How2eval
  9. 参考资料 Reference

性能情况

训练数据集 权值文件名称 测试数据集 输入图片大小 mAP 0.5:0.95 mAP 0.5
COCO-Train2017 yolox_s.pth COCO-Val2017 640x640 38.2 57.7
COCO-Train2017 yolox_m.pth COCO-Val2017 640x640 44.8 63.9
COCO-Train2017 yolox_l.pth COCO-Val2017 640x640 47.9 66.6
COCO-Train2017 yolox_x.pth COCO-Val2017 640x640 49.0 67.7

实现的内容

  • 主干特征提取网络：使用了Focus网络结构。
  • 分类回归层：Decoupled Head，在YoloX中，Yolo Head被分为了分类回归两部分，最后预测的时候才整合在一起。
  • 训练用到的小技巧：Mosaic数据增强、CIOU（原版是IOU和GIOU，CIOU效果类似，都是IOU系列的，甚至更新一些）、学习率余弦退火衰减。
  • Anchor Free：不使用先验框
  • SimOTA：为不同大小的目标动态匹配正样本。

所需环境

pytorch==1.2.0

小技巧的设置

在train.py文件下：
1、mosaic参数可用于控制是否实现Mosaic数据增强。
2、Cosine_scheduler可用于控制是否使用学习率余弦退火衰减。
3、label_smoothing可用于控制是否Label Smoothing平滑。

文件下载

训练所需的权值可在百度网盘中下载。
链接: https://pan.baidu.com/s/1OnM-uWKETFJh_uFCAK6Vlg 提取码: b6km

VOC数据集下载地址如下：
VOC2007+2012训练集
链接: https://pan.baidu.com/s/16pemiBGd-P9q2j7dZKGDFA 提取码: eiw9

VOC2007测试集
链接: https://pan.baidu.com/s/1BnMiFwlNwIWG9gsd4jHLig 提取码: dsda

训练步骤

a、数据集的准备

1、本文使用VOC格式进行训练，训练前需要自己制作好数据集，如果没有自己的数据集，可以通过Github连接下载VOC12+07的数据集尝试下。
2、训练前将标签文件放在VOCdevkit文件夹下的VOC2007文件夹下的Annotation中。
3、训练前将图片文件放在VOCdevkit文件夹下的VOC2007文件夹下的JPEGImages中。

b、数据集的预处理

1、训练数据集时，在model_data文件夹下建立一个cls_classes.txt，里面写所需要区分的类别。
2、设置根目录下的voc_annotation.py里的一些参数。第一次训练可以仅修改classes_path，classes_path用于指向检测类别所对应的txt，即：

classes_path = 'model_data/cls_classes.txt'

model_data/cls_classes.txt文件内容为：

cat
dog
...

3、设置完成后运行voc_annotation.py，生成训练所需的2007_train.txt以及2007_val.txt。

c、开始网络训练

1、通过voc_annotation.py，我们已经生成了2007_train.txt以及2007_val.txt，此时我们可以开始训练了。
2、设置根目录下的train.py里的一些参数。第一次训练可以仅修改classes_path，classes_path用于指向检测类别所对应的txt，设置方式与b、数据集的预处理类似。训练自己的数据集必须要修改！
3、设置完成后运行train.py开始训练了，在训练多个epoch后，权值会生成在logs文件夹中。
4、训练的参数较多，大家可以在下载库后仔细看注释，其中最重要的部分依然是train.py里的classes_path。

d、训练结果预测

1、训练结果预测需要用到两个文件，分别是yolo.py和predict.py。
2、设置根目录下的yolo.py里的一些参数。第一次预测可以仅修改model_path以及classes_path。训练自己的数据集必须要修改。model_path指向训练好的权值文件，在logs文件夹里。classes_path指向检测类别所对应的txt。
3、设置完成后运行predict.py开始预测了，具体细节查看预测步骤。
4、预测的参数较多，大家可以在下载库后仔细看注释，其中最重要的部分依然是yolo.py里的model_path以及classes_path。

预测步骤

a、使用预训练权重

1、下载完库后解压，在百度网盘下载各个权值，放入model_data，默认使用yolox_s.pth，其它可调整，运行predict.py，输入

img/street.jpg

2、在predict.py里面进行设置可以进行video视频检测、fps测试、批量文件测试与保存。

b、使用自己训练的权重

1、按照训练步骤训练。
2、在yolo.py文件里面，在如下部分修改model_path和classes_path使其对应训练好的文件；model_path对应logs文件夹下面的权值文件，classes_path是model_path对应分的类。

_defaults = {
    #--------------------------------------------------------------------------#
    #   使用自己训练好的模型进行预测一定要修改model_path和classes_path！
    #   model_path指向logs文件夹下的权值文件，classes_path指向model_data下的txt
    #   如果出现shape不匹配，同时要注意训练时的model_path和classes_path参数的修改
    #--------------------------------------------------------------------------#
    "model_path"        : 'model_data/yolox_s.pth',
    "classes_path"      : 'model_data/coco_classes.txt',
    #---------------------------------------------------------------------#
    #   输入图片的大小，必须为32的倍数。
    #---------------------------------------------------------------------#
    "input_shape"       : [640, 640],
    #---------------------------------------------------------------------#
    #   所使用的YoloX的版本。s、m、l、x
    #---------------------------------------------------------------------#
    "phi"               : 's',
    #---------------------------------------------------------------------#
    #   只有得分大于置信度的预测框会被保留下来
    #---------------------------------------------------------------------#
    "confidence"        : 0.5,
    #---------------------------------------------------------------------#
    #   非极大抑制所用到的nms_iou大小
    #---------------------------------------------------------------------#
    "nms_iou"           : 0.3,
    #---------------------------------------------------------------------#
    #   该变量用于控制是否使用letterbox_image对输入图像进行不失真的resize，
    #   在多次测试后，发现关闭letterbox_image直接resize的效果更好
    #---------------------------------------------------------------------#
    "letterbox_image"   : True,
    #-------------------------------#
    #   是否使用Cuda
    #   没有GPU可以设置成False
    #-------------------------------#
    "cuda"              : True,
}

3、运行predict.py，输入

img/street.jpg

4、在predict.py里面进行设置可以进行video视频检测、fps测试、批量文件测试与保存。

评估步骤

1、本文使用VOC格式进行评估。
2、划分测试集，如果在训练前已经运行过voc_annotation.py文件，代码会自动将数据集划分成训练集、验证集和测试集。
3、如果想要修改测试集的比例，可以修改voc_annotation.py文件下的trainval_percent。trainval_percent用于指定(训练集+验证集)与测试集的比例，默认情况下 (训练集+验证集):测试集 = 9:1。train_percent用于指定(训练集+验证集)中训练集与验证集的比例，默认情况下 训练集:验证集 = 9:1。
4、设置根目录下的yolo.py里的一些参数。第一次评估可以仅修改model_path以及classes_path。训练自己的数据集必须要修改。model_path指向训练好的权值文件，在logs文件夹里。classes_path指向检测类别所对应的txt。
5、设置根目录下的get_map.py里的一些参数。第一次评估可以仅修改classes_path，classes_path用于指向检测类别所对应的txt，评估自己的数据集必须要修改。与yolo.py中分开设置的原因是可以让使用者自己选择评估什么类别，而非所有类别。
6、运行get_map.py即可获得评估结果，评估结果会保存在map_out文件夹中。

Reference

https://github.com/Megvii-BaseDetection/YOLOX

Comments
  • 在使用YOLOX模型 对视频进行预测时，出现了如下错误

    在使用YOLOX模型 对视频进行预测时，出现了如下错误

    在使用YOLOX模型 对视频进行预测时，出现了如下错误： cv2.error: OpenCV(4.5.4-dev) D:\a\opencv-python\opencv-python\opencv\modules\imgproc\src\color.cpp:182: error: (-215:Assertion failed) !_src.empty() in function 'cv::cvtColor' image

    请问如何解决呀？

    opened by MasterMiao919 7
  • 训练自己数据，MAP出现问题

    训练自己数据，MAP出现问题

    hi，博主 关于训练自己的数据集， 已经将对应格式的文件放到相同路径下的文件夹内，新添了自己的cls.txt。训练完成后，也有训练框。 但是，测试map时，仍然有原voc的测试类别，想问一下这是什么情况呢？ （ classes_path = 'model_data/fire.txt'已经就改）

    opened by theDeep1nteresting 3
  • 网络输出的代码错了吧？output = torch.cat([reg_output, obj_output, cls_output], 1)

    网络输出的代码错了吧？output = torch.cat([reg_output, obj_output, cls_output], 1)

    复现代码是 output = torch.cat([reg_output, obj_output, cls_output], 1) 源代码是 output = torch.cat( [reg_output, obj_output.sigmoid(), cls_output.sigmoid()], 1 ) 复现代码没加激活函数啊？

    opened by mepleleo 1
  • ModuleNotFoundError: No module named 'models'

    ModuleNotFoundError: No module named 'models'

    Traceback (most recent call last): File "O:\graduate\yolov7-bubbliiiing\predict.py", line 15, in yolo = YOLO() File "O:\graduate\yolov7-bubbliiiing\yolo.py", line 95, in init self.generate() File "O:\graduate\yolov7-bubbliiiing\yolo.py", line 108, in generate self.net.load_state_dict(torch.load(self.model_path, map_location=device)) File "D:\Anaconda\envs\pytorch-gpu\lib\site-packages\torch\serialization.py", line 592, in load return _load(opened_zipfile, map_location, pickle_module, **pickle_load_args) File "D:\Anaconda\envs\pytorch-gpu\lib\site-packages\torch\serialization.py", line 851, in _load result = unpickler.load() ModuleNotFoundError: No module named 'models'

    这是跑大佬您的yolov7时运行predict.py的问题，发错地方了😂

    opened by lip111 1
  • involution卷积替换问题

    involution卷积替换问题

    up，最近看了一些论文显示involution卷积效果不错，想来替换试试，但是involution官方代码，参数和yolox的不太匹配，调整了好久都一直报错，能麻烦up指点一下 参数该如何修改呢qwq？ import torch.nn as nn from mmcv.cnn import ConvModule

    class involution(nn.Module):

    def __init__(self,
             channels,
             kernel_size,
             stride):
    super(involution, self).__init__()
    self.kernel_size = kernel_size
    self.stride = stride
    self.channels = channels
    reduction_ratio = 4
    self.group_channels = 16
    self.groups = self.channels // self.group_channels
    self.conv1 = ConvModule(
        in_channels=channels,
        out_channels=channels // reduction_ratio,
        kernel_size=1,
        conv_cfg=None,
        norm_cfg=dict(type='BN'),
        act_cfg=dict(type='ReLU'))
    self.conv2 = ConvModule(
        in_channels=channels // reduction_ratio,
        out_channels=kernel_size**2 * self.groups,
        kernel_size=1,
        stride=1,
        conv_cfg=None,
        norm_cfg=None,
        act_cfg=None)
    if stride > 1:
        self.avgpool = nn.AvgPool2d(stride, stride)
    self.unfold = nn.Unfold(kernel_size, 1, (kernel_size-1)//2, stride)

def forward(self, x):
    weight = self.conv2(self.conv1(x if self.stride == 1 else self.avgpool(x)))
    b, c, h, w = weight.shape
    weight = weight.view(b, self.groups, self.kernel_size**2, h, w).unsqueeze(2)
    out = self.unfold(x).view(b, self.groups, self.group_channels, self.kernel_size**2, h, w)
    out = (weight * out).sum(dim=3).view(b, self.channels, h, w)
    return out
    opened by right135 3
Releases(v2.1)
Owner
Bubbliiiing
Bubbliiiing
GitHub Repository
Hybrid Neural Fusion for Full-frame Video Stabilization

FuSta: Hybrid Neural Fusion for Full-frame Video Stabilization Project Page | Video | Paper | Google Colab Setup Setup environment for [Yu and Ramamoo

Yu-Lun Liu 430 Jan 04, 2023
Snapchat-filters-app-opencv-python - Here we used opencv and other inbuilt python modules to create filter application like snapchat

Snapchat like filter App using opencv python Backend : opencv and python Library

2 Jul 19, 2022
TAUFE: Task-Agnostic Undesirable Feature DeactivationUsing Out-of-Distribution Data

A deep neural network (DNN) has achieved great success in many machine learning tasks by virtue of its high expressive power. However, its prediction can be easily biased to undesirable features, whi

KAIST Data Mining Lab 8 Dec 07, 2022
SuMa++: Efficient LiDAR-based Semantic SLAM (Chen et al IROS 2019)

SuMa++: Efficient LiDAR-based Semantic SLAM This repository contains the implementation of SuMa++, which generates semantic maps only using three-dime

Photogrammetry & Robotics Bonn 701 Dec 30, 2022
People movement type classifier with YOLOv4 detection and SORT tracking.

Movement classification The goal of this project would be movement classification of people, in other words, walking (normal and fast) and running. Yo

4 Sep 21, 2021
Neural Scene Flow Fields using pytorch-lightning, with potential improvements

nsff_pl Neural Scene Flow Fields using pytorch-lightning. This repo reimplements the NSFF idea, but modifies several operations based on observation o

AI葵 178 Dec 21, 2022
An All-MLP solution for Vision, from Google AI

MLP Mixer - Pytorch An All-MLP solution for Vision, from Google AI, in Pytorch. No convolutions nor attention needed! Yannic Kilcher video Install $ p

Phil Wang 784 Jan 06, 2023
Official implementation of Self-supervised Graph Attention Networks (SuperGAT), ICLR 2021.

SuperGAT Official implementation of Self-supervised Graph Attention Networks (SuperGAT). This model is presented at How to Find Your Friendly Neighbor

Dongkwan Kim 127 Dec 28, 2022
A high-performance distributed deep learning system targeting large-scale and automated distributed training.

HETU Documentation | Examples Hetu is a high-performance distributed deep learning system targeting trillions of parameters DL model training, develop

DAIR Lab 150 Dec 21, 2022
[EMNLP 2020] Keep CALM and Explore: Language Models for Action Generation in Text-based Games

Contextual Action Language Model (CALM) and the ClubFloyd Dataset Code and data for paper Keep CALM and Explore: Language Models for Action Generation

Princeton Natural Language Processing 43 Dec 16, 2022
Source code for CVPR 2020 paper "Learning to Forget for Meta-Learning"

L2F - Learning to Forget for Meta-Learning Sungyong Baik, Seokil Hong, Kyoung Mu Lee Source code for CVPR 2020 paper "Learning to Forget for Meta-Lear

Sungyong Baik 29 May 22, 2022
Content shared at DS-OX Meetup

Streamlit-Projects Streamlit projects available in this repo: An introduction to Streamlit presented at DS-OX (Feb 26, 2020) meetup Streamlit 101 - Ja

Arvindra 69 Dec 23, 2022
Optimizing DR with hard negatives and achieving SOTA first-stage retrieval performance on TREC DL Track (SIGIR 2021 Full Paper).

Optimizing Dense Retrieval Model Training with Hard Negatives Jingtao Zhan, Jiaxin Mao, Yiqun Liu, Jiafeng Guo, Min Zhang, Shaoping Ma 🔥 News 2021-10

Jingtao Zhan 99 Dec 27, 2022
Multi-Joint dynamics with Contact. A general purpose physics simulator.

MuJoCo Physics MuJoCo stands for Multi-Joint dynamics with Contact. It is a general purpose physics engine that aims to facilitate research and develo

DeepMind 5.2k Jan 02, 2023
Accelerated deep learning R&D

Accelerated deep learning R&D PyTorch framework for Deep Learning research and development. It focuses on reproducibility, rapid experimentation, and

Catalyst-Team 3.1k Jan 06, 2023
VM3000 Microphones

VM3000-Microphones This project was completed by Ricky Leman under the supervision of Dr Ben Travaglione and Professor Melinda Hodkiewicz as part of t

UWA System Health Lab 0 Jun 04, 2021
Speeding-Up Back-Propagation in DNN: Approximate Outer Product with Memory

Approximate Outer Product Gradient Descent with Memory Code for the numerical experiment of the paper Speeding-Up Back-Propagation in DNN: Approximate

2 Mar 02, 2022
A python script to dump all the challenges locally of a CTFd-based Capture the Flag.

A python script to dump all the challenges locally of a CTFd-based Capture the Flag. Features Connects and logins to a remote CTFd instance. Dumps all

Podalirius 77 Dec 07, 2022
Plug-n-Play Reinforcement Learning in Python with OpenAI Gym and JAX

coax is built on top of JAX, but it doesn't have an explicit dependence on the jax python package. The reason is that your version of jaxlib will depend on your CUDA version.

128 Dec 27, 2022
AI-based, context-driven network device ranking

Batea A batea is a large shallow pan of wood or iron traditionally used by gold prospectors for washing sand and gravel to recover gold nuggets. Batea

Secureworks Taegis VDR 269 Nov 26, 2022