# YOLOv4 Training Tutorial
[TOC]
## 0. First of all
This tutorial will help you build YOLOv4 easily in the cloud with GPU enabled so that you can run object detections in milliseconds!
https://colab.research.google.com/drive/1_GdoqCJWXsChrOiY8sZMr_zbr_fH-0Fg?usp=sharing&fbclid=IwAR0DRwtH6-9D_gYbqxpCFdRIeBYJ1ozle0AiAU3hxk3gOjSePSMbc-JhBaY#scrollTo=q2Jjv0yRKLPe
## 1. Git Clone Darknet
```cmd
git clone https://github.com/AlexeyAB/darknet
```
## 2. Change makefile to have GPU and OPENCV enabled
```cmd
cd darknet
sed -i 's/OPENCV=0/OPENCV=1/' Makefile
sed -i 's/GPU=0/GPU=1/' Makefile
sed -i 's/CUDNN=0/CUDNN=1/' Makefile
sed -i 's/CUDNN_HALF=0/CUDNN_HALF=1/' Makefile
```
verify CUDA
```cmd
!/usr/local/cuda/bin/nvcc --version
```
==Notice that the version of cuda needs >= 10.2==
> If you don't know how to establish cuda & cudnn in Ubuntu environment.
Please refer to https://medium.com/ching-i/ubuntu-%E5%AE%89%E8%A3%9D-gpu-driver-cuda-cudnn-%E6%95%99%E5%AD%B8-972b189a6d35
## 3. Make Darknet
make darknet, builds darknet so that you can then use the darknet executable file to run or train object detectors
```cmd
make
```
## 4. Files Architecture
#### Archive
```
./darknet
| |－generate_test.py
| |－generate_train.py
| |－data
| | |－obj
| | | |－0000000.png
| | | |－0000000.txt
| | | |－...
| | |－test
| | | |－1000001.png
| | | |－1000001.txt
| | | |－...
| | |－obj.names
| | |－obj.data
| |－cfg
| | |－yolov4-obj.cfg
```
#### obj.data
```cmd
classes = 3
train = /home/chenging/Autonomous_Driving_Final/darknet/data/train.txt
valid = /home/chenging/Autonomous_Driving_Final/darknet/data/test.txt
names = /home/chenging/Autonomous_Driving_Final/darknet/data/obj.names
backup = /home/chenging/Autonomous_Driving_Final/darknet/backup
```
https://drive.google.com/file/d/1oVbwfLfsyttismgoRmRHM35WGqDUckCa/view?usp=sharing
#### obj.names
```cmd
Car
Pedestrian
Cyclist
```
https://drive.google.com/file/d/1mKFsJ3wdBU0I2fXZ_a4vrQeEIrqG3z9x/view?usp=sharing
#### yolov4-obj.cfg
```
[net]
# Testing # If you've trained, please un-command it.
# batch=1
# subdivisions=1
Training
batch=64
subdivisions=16 # if your memory is not enough, please increase it!
width=416
height=416
channels=3
momentum=0.949
decay=0.0005
angle=0
saturation = 1.5
exposure = 1.5
hue=.1
learning_rate=0.001
burn_in=1000
max_batches = 6000
policy=steps
steps=4800,5400
scales=.1,.1
```
https://drive.google.com/file/d/1H1uVhalKED9bQmVxSQQhEwBScM2zOjti/view?usp=sharing
You can find more details about yolov4-obj.cfg in the following website :
https://codertw.com/%E7%A8%8B%E5%BC%8F%E8%AA%9E%E8%A8%80/542552/
#### 0000000.png
#### 0000000.txt
```cmd
0 0.570234375 0.389734375 0.11641145833333327 0.1913125
0 0.6023385416666667 0.4096484375 0.05925 0.09820312499999995
0 0.7507083333333333 0.5038671875 0.2666354166666667 0.1546796875
0 0.570234375 0.389734375 0.11641145833333327 0.1913125
0 0.6023385416666667 0.4096484375 0.05925 0.09820312499999995
0 0.7507083333333333 0.5038671875 0.2666354166666667 0.1546796875
```
* 0000000.txt needs follow YOLO Format
Source : https://chtseng.wordpress.com/2018/09/01/%E5%BB%BA%E7%AB%8B%E8%87%AA%E5%B7%B1%E7%9A%84yolo%E8%BE%A8%E8%AD%98%E6%A8%A1%E5%9E%8B-%E4%BB%A5%E6%9F%91%E6%A9%98%E8%BE%A8%E8%AD%98%E7%82%BA%E4%BE%8B/
> What if my format isn't yolo format but waymo
> Please refer to
> 1. https://github.com/caizhongang/waymo_kitti_converter.git
> 2. https://github.com/EscVM/OIDv4_ToolKit
## 5. Pre-trained weight
```cmd
wget https://github.com/AlexeyAB/darknet/releases/download/darknet_yolo_v3_optimal/yolov4.conv.137
```
## 6. Train it and Demo!
```cmd
./darknet detector train data/obj.data cfg/yolov4-obj.cfg yolov4.conv.137 -dont_show -map
```
```cmd
./darknet detector train data/obj.data cfg/yolov4-obj.cfg backup/yolov4-obj_best.weights -dont_show -map
```
If you have any pre-trained weight, please replace backup/yolov4-obj_best.weights or yolov4.conv.137 with your pre-trained weight.
In ./darknet you can see the training process plot if you add -map. Every 100 iterations, it will compute mAP.
> What is mAP?
https://chih-sheng-huang821.medium.com/%E6%B7%B1%E5%BA%A6%E5%AD%B8%E7%BF%92%E7%B3%BB%E5%88%97-%E4%BB%80%E9%BA%BC%E6%98%AFap-map-aaf089920848?fbclid=IwAR2TqOW3XAA7TTG6efarRTdVPuQgxn_c8VDzHeRtJ9RRPm__y8Y6nc19TH0
#### Test a image
```
./darknet detector test data/obj.data cfg/yolov4-obj.cfg ./backup/yolov4-obj_best.weights ./test.png -thresh 0.3 -out result.txt
```
```
[
{
"frame_id":1,
"filename":"./test.png",
"objects": [
{"class_id":1, "name":"Pedestrian", "relative_coordinates":{"center_x":0.025277, "center_y":0.590687, "width":0.056367, "height":0.105150}, "confidence":0.974176},
{"class_id":0, "name":"Car", "relative_coordinates":{"center_x":0.379112, "center_y":0.568891, "width":0.094814, "height":0.111459}, "confidence":0.993241},
{"class_id":0, "name":"Car", "relative_coordinates":{"center_x":0.809745, "center_y":0.620679, "width":0.186627, "height":0.168321}, "confidence":0.989374},
{"class_id":0, "name":"Car", "relative_coordinates":{"center_x":0.650084, "center_y":0.541073, "width":0.065840, "height":0.081862}, "confidence":0.984396},
{"class_id":0, "name":"Car", "relative_coordinates":{"center_x":0.502944, "center_y":0.527315, "width":0.068330, "height":0.106554}, "confidence":0.982795},
{"class_id":0, "name":"Car", "relative_coordinates":{"center_x":0.623770, "center_y":0.525190, "width":0.061993, "height":0.068749}, "confidence":0.968530},
{"class_id":0, "name":"Car", "relative_coordinates":{"center_x":0.272871, "center_y":0.559579, "width":0.083208, "height":0.074861}, "confidence":0.967086},
{"class_id":0, "name":"Car", "relative_coordinates":{"center_x":0.339403, "center_y":0.532364, "width":0.062299, "height":0.064379}, "confidence":0.933663},
{"class_id":0, "name":"Car", "relative_coordinates":{"center_x":0.301758, "center_y":0.547045, "width":0.063640, "height":0.067330}, "confidence":0.930660},
{"class_id":0, "name":"Car", "relative_coordinates":{"center_x":0.432346, "center_y":0.529242, "width":0.058491, "height":0.070736}, "confidence":0.923266},
{"class_id":0, "name":"Car", "relative_coordinates":{"center_x":0.605002, "center_y":0.517186, "width":0.048999, "height":0.064098}, "confidence":0.906764},
{"class_id":0, "name":"Car", "relative_coordinates":{"center_x":0.227240, "center_y":0.579499, "width":0.114512, "height":0.085782}, "confidence":0.887934},
{"class_id":0, "name":"Car", "relative_coordinates":{"center_x":0.368436, "center_y":0.531378, "width":0.060252, "height":0.070806}, "confidence":0.776398},
{"class_id":0, "name":"Car", "relative_coordinates":{"center_x":0.589318, "center_y":0.511667, "width":0.038385, "height":0.052332}, "confidence":0.403203},
{"class_id":0, "name":"Car", "relative_coordinates":{"center_x":0.315539, "center_y":0.540429, "width":0.060694, "height":0.062630}, "confidence":0.303930}
]
}
]
```
#### Test images
```
/darknet detector test data/obj.data cfg/yolov4-obj.cfg ./backup/yolov4-obj_best.weights -thresh 0.25 -ext_output <Test/Test.txt> result.txt
```
#### Test.txt
```
./Test/1001000.png
./Test/1001001.png
./Test/1001002.png
./Test/1001003.png
./Test/1001004.png
./Test/1001005.png
./Test/1001006.png
./Test/1001007.png
./Test/1001008.png
./Test/1001009.png
./Test/1001010.png
./Test/1001011.png
./Test/1001012.png
./Test/1001013.png
./Test/1001014.png
./Test/1001015.png
./Test/1001016.png
./Test/1001017.png
./Test/1001018.png
./Test/1001019.png
```
#### result.txt
```
CUDNN_HALF=1
net.optimized_memory = 0
mini_batch = 1, batch = 1, time_steps = 1, train = 0
Create CUDA-stream - 0
Create cudnn-handle 0
nms_kind: greedynms (1), beta = 0.600000
nms_kind: greedynms (1), beta = 0.600000
nms_kind: greedynms (1), beta = 0.600000
seen 64, trained: 131 K-images (2 Kilo-batches_64)
Enter Image Path: Detection layer: 139 - type = 28
Detection layer: 150 - type = 28
Detection layer: 161 - type = 28
./Test/1001000.png: Predicted in 133.273000 milli-seconds.
Car: 98% (left_x: 1 top_y: 741 width: 93 height: 150)
Car: 89% (left_x: 23 top_y: 733 width: 178 height: 100)
Car: 98% (left_x: 312 top_y: 724 width: 222 height: 138)
Car: 99% (left_x: 560 top_y: 708 width: 225 height: 198)
Car: 79% (left_x: 653 top_y: 703 width: 147 height: 115)
Car: 99% (left_x: 1500 top_y: 713 width: 422 height: 275)
Enter Image Path: Detection layer: 139 - type = 28
Detection layer: 150 - type = 28
Detection layer: 161 - type = 28
./Test/1001001.png: Predicted in 116.374000 milli-seconds.
Car: 97% (left_x: 1 top_y: 740 width: 91 height: 155)
Car: 90% (left_x: 22 top_y: 735 width: 179 height: 100)
Car: 97% (left_x: 310 top_y: 722 width: 225 height: 138)
Car: 99% (left_x: 564 top_y: 707 width: 217 height: 204)
Car: 84% (left_x: 651 top_y: 701 width: 146 height: 116)
Car: 99% (left_x: 1433 top_y: 715 width: 493 height: 276)
Enter Image Path: Detection layer: 139 - type = 28
Detection layer: 150 - type = 28
Detection layer: 161 - type = 28
./Test/1001002.png: Predicted in 117.112000 milli-seconds.
Car: 97% (left_x: 2 top_y: 738 width: 91 height: 158)
Car: 85% (left_x: 23 top_y: 734 width: 178 height: 101)
Car: 98% (left_x: 312 top_y: 720 width: 217 height: 143)
Car: 99% (left_x: 559 top_y: 708 width: 229 height: 193)
Car: 86% (left_x: 649 top_y: 698 width: 151 height: 115)
Car: 98% (left_x: 1369 top_y: 711 width: 558 height: 276)
...
```
#### Test video
```
./darknet detector demo data/obj.data cfg/yolov4-obj.cfg ./backup/yolov4-obj_best.weights -thresh 0.25 -dont_show -ext_output ./test.mp4 -out_filename result.avi
```
https://www.youtube.com/watch?v=DCLd-WqeBcY
Sign in with Wallet
Connect another wallet