Underwater industrial application yolov5m6

Overview

underwater-industrial-application-yolov5m6

This project wins the intelligent algorithm contest finalist award and stands out from over 2000teams in China Underwater Robot Professional Contest, entering the final of China Underwater Robot Professional Contest and ranking 13 out of 31 teams in finals.

和鲸社区Kesci 水下光学目标检测产业应用赛项

环境:

mmdetection

+ 操作系统:Ubuntu 18.04.2
+ GPU:1块2080Ti
+ Python:Python 3.7.7
+ NVIDIA依赖:
    - NVCC: Cuda compilation tools, release 10.1, V10.1.243
    - CuDNN 7.6.5
+ 深度学习框架:
    - PyTorch: 1.8.1
    - TorchVision: 0.9.1
    - OpenCV
    - MMCV
    - MMDetection(注意data clean 的版本不同)

yolov5

训练环境:
	+ 操作系统:Ubuntu 18.04.2
	+ GPU:1块2080Ti
	+ Python:Python 3.7.7
测试环境:
	 NVIDIA Jetson AGX Xavier


# pip install -r requirements.txt

# base ----------------------------------------
matplotlib>=3.2.2
numpy>=1.18.5
opencv-python>=4.1.2
Pillow
PyYAML>=5.3.1
scipy>=1.4.1
torch>=1.7.0
torchvision>=0.8.1
tqdm>=4.41.0

# logging -------------------------------------
tensorboard>=2.4.1
# wandb

# plotting ------------------------------------
seaborn>=0.11.0
pandas

# export --------------------------------------
# coremltools>=4.1
# onnx>=1.9.0
# scikit-learn==0.19.2  # for coreml quantization
# tensorflow==2.4.1  # for TFLite export

# extras --------------------------------------
# Cython  # for pycocotools https://github.com/cocodataset/cocoapi/issues/172
# pycocotools>=2.0  # COCO mAP
# albumentations>=1.0.3
thop  # FLOPs computation

第一大步:@数据清理

文件说明:data_clean_Code用于数据清理

data_clean_Code/yangtiming-underwater-master ->为湛江赛拿第20名方案
data_clean_Code/underwater-detection-master  ->为triks团队湛江赛方案

使用说明

1. (这一步用我的yangtiming-underwater-master替代原有的cascade_rcnn_x101_64x4d_fpn_dcn_e15 )【原因精度更高A榜0.562】

模型采用 cascade_rcnn_x101_64x4d_fpn_dcn_e15  
+ Backbone:
    + ResNeXt101-64x4d
+ Neck:
    + FPN
+ DCN
+ Global context(GC)
+ MS [(4096, 600), (4096, 1000)]
+ RandomRotate90°
+ 15epochs + step:[11, 13]  
+ A榜:0.55040585 
    + 注:不是所有数据

2. 基于1训练好的模型对训练数据进行清洗(tools/data_process/data_clean.py)

+ 1. 如果某张图片上所有预测框的confidence没有一个是大于0.9, 那么去掉该图片(即看不清的图片)
+ 2. 修正错误标注
    + 1. 先过滤掉confidence<0.1的predict boxes, 然后同GT boxes求iou
    + 2. 如果predict box同GT的最大iou大于0.6,但类别不一致, 那么就修正该gt box的类别
trainall.json (与bbox1)修后的到   trainall-revised.json

3. 基于2修正后的数据进行训练->(基于2修正后的到 trainall-revised.json 修正采用cascade_rcnn_r50_rfp_sac后的到-> bbox3

模型采用cascade_rcnn_r50_rfp_sac
+ Backbone:
+ ResNet50
+ Neck:
RFP-SAC
+ GC + MS + RandomRotate90°
+ cascade_iou调整为:(0.55, 0.65, 0.75)
+ A榜: 0.56339531
+ 注:所有数据

4. 基于3训练好的模型进一步清洗数据.

->  trainall-revised-v3.json(与bbox3) 	进一步清洗数据 -> trainall-revised-v4.json)
+ 模型同3: 
+ A榜:0.56945031
    + 注:所有数据
在验证集上面测试精度:1. 执行完mAP0.5:0.95=0.547 4. 执行完mAP0.5:0.95 = 0.560

第二大步:@数据清理完毕后,改用yolov5 (code/yolov5_V5_chuli_focal_loss_attention)

使用背景介绍:
使用模型为yolov5m6系列,迭代tricks的时候,采取用--img 640 进行迭代

最优模型:

最终在yolov5m6上面的精度为:mAP0.5:0.95= 0.5374,agx速度0.2s每张
最好模型:
1.yolov5m6
2.数据清洗
2.attention模块:senet
3.hsv_h,hsv_s,hsv_v=0
4.focal_loss

使用的tricks如下:(按照时间顺序)

1.按照第一大步的数据清洗:由原来的mAP0.5:0.95= 0.465->0.4880
2.yolov5当中的hsv_h,hsv_s,hsv_v均设为0,mAP0.5:0.95= 0.4880 ->0.4940
3.在loss.py当中加入focal_loss损失函数(157行,172行),mAP0.5:0.95= 0.4940 ->0.4977
4.更改原有yolov5的c3层改为senet(attention模块),mAP0.5:0.95= 0.4977 -> 0.50069

以上按照

python train.py  --weights weights/yolov5m6.pt --cfg models/hub/yolov5m6-senet.yaml --data data/underwater.yaml  --hyp data/hyps/hyp.scratch-p6.yaml --epochs 100 --batch-size 25 --img 640

最终要提交的时候,按照

python train.py  --weights weights/yolov5m6.pt --cfg models/hub/yolov5m6-senet.yaml --data data/underwater.yaml  --hyp data/hyps/hyp.scratch-p6.yaml --epochs 250 --batch-size 4 --img 1280 --multi-scale

【注意:multi-scale大小可以在train.py文件夹下面更改】

测试

python3 val_tm_txt_csv.py --data  /data/underwater.yaml   --weights weights/best_05374.pt --img 1280 --save-txt --save-conf --half

【--half能提升速度(fp16),精度比fp32更高】

################

若要测试mAP,可以用 https://github.com/rafaelpadilla/review_object_detection_metrics.git

以下为比赛文档说明

若有权限问题,则执行前 chmod +x main_test.sh

1. 将测试集的图片放在本文件夹当中名字为test
2.最终结果将放在answer当中(执行后自动生成)
3.code文件夹为全部代码,以及包含训练权重
4.执行main_test.sh开始运行



(*)Q:何时开始计时?(注意:在执行main_test.sh之前命令窗口拉长,否则计时无法看到进度条)
当执行 python3 ./val_tm_txt_csv.py 时,看见如下界面表示计时开始
##                 Class     Images     Labels          P          R     [email protected] [email protected]:.95:   0%|          | 0/xxx [00:00

reference

+yolov5

+yangtiming/underwater-mmdetection

+team-tricks

Instance-based label smoothing for improving deep neural networks generalization and calibration

Instance-based Label Smoothing for Neural Networks Pytorch Implementation of the algorithm. This repository includes a new proposed method for instanc

Mohamed Maher 1 Aug 13, 2022
[NeurIPS'21] Shape As Points: A Differentiable Poisson Solver

Shape As Points (SAP) Paper | Project Page | Short Video (6 min) | Long Video (12 min) This repository contains the implementation of the paper: Shape

394 Dec 30, 2022
DeepFaceLive - Live Deep Fake in python, Real-time face swap for PC streaming or video calls

DeepFaceLive - Live Deep Fake in python, Real-time face swap for PC streaming or video calls

8.3k Dec 31, 2022
TimeSHAP explains Recurrent Neural Network predictions.

TimeSHAP TimeSHAP is a model-agnostic, recurrent explainer that builds upon KernelSHAP and extends it to the sequential domain. TimeSHAP computes even

Feedzai 90 Dec 18, 2022
MOOSE (Multi-organ objective segmentation) a data-centric AI solution that generates multilabel organ segmentations to facilitate systemic TB whole-person research

MOOSE (Multi-organ objective segmentation) a data-centric AI solution that generates multilabel organ segmentations to facilitate systemic TB whole-person research.The pipeline is based on nn-UNet an

QIMP team 30 Jan 01, 2023
This repository contains the scripts for downloading and validating scripts for the documents

HC4: HLTCOE CLIR Common-Crawl Collection This repository contains the scripts for downloading and validating scripts for the documents. Document ids,

JHU Human Language Technology Center of Excellence 6 Jun 07, 2022
git《Investigating Loss Functions for Extreme Super-Resolution》(CVPR 2020) GitHub:

Investigating Loss Functions for Extreme Super-Resolution NTIRE 2020 Perceptual Extreme Super-Resolution Submission. Our method ranked first and secon

Sejong Yang 0 Oct 17, 2022
Code for Iso-Points: Optimizing Neural Implicit Surfaces with Hybrid Representations

Implementation for Iso-Points (CVPR 2021) Official code for paper Iso-Points: Optimizing Neural Implicit Surfaces with Hybrid Representations paper |

Yifan Wang 66 Nov 08, 2022
Implementation of "Learning Multi-Granular Hypergraphs for Video-Based Person Re-Identification"

hypergraph_reid Implementation of "Learning Multi-Granular Hypergraphs for Video-Based Person Re-Identification" If you find this help your research,

62 Dec 21, 2022
A Tensorfflow implementation of Attend, Infer, Repeat

Attend, Infer, Repeat: Fast Scene Understanding with Generative Models This is an unofficial Tensorflow implementation of Attend, Infear, Repeat (AIR)

Adam Kosiorek 82 May 27, 2022
This repository allows the user to automatically scale a 3D model/mesh/point cloud on Agisoft Metashape

Metashape-Utils This repository allows the user to automatically scale a 3D model/mesh/point cloud on Agisoft Metashape, given a set of 2D coordinates

INSCRIBE 4 Nov 07, 2022
[NeurIPS 2021 Spotlight] Aligning Pretraining for Detection via Object-Level Contrastive Learning

SoCo [NeurIPS 2021 Spotlight] Aligning Pretraining for Detection via Object-Level Contrastive Learning By Fangyun Wei*, Yue Gao*, Zhirong Wu, Han Hu,

Yue Gao 139 Dec 14, 2022
(ICCV 2021) Official code of "Dressing in Order: Recurrent Person Image Generation for Pose Transfer, Virtual Try-on and Outfit Editing."

Dressing in Order (DiOr) 👚 [Paper] 👖 [Webpage] 👗 [Running this code] The official implementation of "Dressing in Order: Recurrent Person Image Gene

Aiyu Cui 277 Dec 28, 2022
Tacotron 2 - PyTorch implementation with faster-than-realtime inference

Tacotron 2 (without wavenet) PyTorch implementation of Natural TTS Synthesis By Conditioning Wavenet On Mel Spectrogram Predictions. This implementati

NVIDIA Corporation 4.1k Jan 03, 2023
Authors implementation of LieTransformer: Equivariant Self-Attention for Lie Groups

LieTransformer This repository contains the implementation of the LieTransformer used for experiments in the paper LieTransformer: Equivariant self-at

35 Oct 18, 2022
This is a simple face recognition mini project that was completed by a team of 3 members in 1 week's time

PeekingDuckling 1. Description This is an implementation of facial identification algorithm to detect and identify the faces of the 3 team members Cla

Eric Kwok 2 Jan 25, 2022
Metrics to evaluate quality and efficacy of synthetic datasets.

An Open Source Project from the Data to AI Lab, at MIT Metrics for Synthetic Data Generation Projects Website: https://sdv.dev Documentation: https://

The Synthetic Data Vault Project 129 Jan 03, 2023
Code in conjunction with the publication 'Contrastive Representation Learning for Hand Shape Estimation'

HanCo Dataset & Contrastive Representation Learning for Hand Shape Estimation Code in conjunction with the publication: Contrastive Representation Lea

Computer Vision Group, Albert-Ludwigs-Universität Freiburg 38 Dec 13, 2022
For the paper entitled ''A Case Study and Qualitative Analysis of Simple Cross-Lingual Opinion Mining''

Summary This is the source code for the paper "A Case Study and Qualitative Analysis of Simple Cross-Lingual Opinion Mining", which was accepted as fu

1 Nov 10, 2021
Algorithmic Trading using RNN

Deep-Trading This an implementation adapted from Rachnog Neural networks for algorithmic trading. Part One — Simple time series forecasting and this c

Hazem Nomer 29 Sep 04, 2022