Final project of CISC 642

Introduction

This repo contains drive code of a recently released model, i.e., YOLOv4.

Rescription

Original repo is referred to https://github.com/AlexeyAB/darknet. The author mainly experiments on MSCOCO. But this repo mainly experiments on VOC dataset.

Requirements

1. Window/Linux
2. CMake >= 3.12: https://cmake.org/download/
3. CUDA 10.0: https://developer.nvidia.com/cuda-toolkit-archive (on Linux do Post-installation Actions)
4. OpenCV >= 2.4: use your preferred package manager (brew, apt), build from source using vcpkg or download from OpenCV official site (on Windows set system variable OpenCV_DIR = C:\opencv\build - where are the include and x64 folders image)
5. cuDNN >= 7.0 for CUDA 10.0 https://developer.nvidia.com/rdp/cudnn-archive (on Linux copy cudnn.h,libcudnn.so... as desribed here https://docs.nvidia.com/deeplearning/sdk/cudnn-install/index.html#installlinux-tar , on Windows copy cudnn.h,cudnn64_7.dll, cudnn64_7.lib as desribed here https://docs.nvidia.com/deeplearning/sdk/cudnn-install/index.html#installwindows )
6. GPU with CC >= 3.0: https://en.wikipedia.org/wiki/CUDA#GPUs_supported
7. on Linux GCC or Clang, on Windows MSVC 2015/2017/2019 https://visualstudio.microsoft.com/thank-you-downloading-visual-studio/?sku=Community

Dataset

download VOC2007 + 2012

$ sh data/VOC2007.sh
$ sh data/VOC2012.sh

And data will placed in data/.

python Then label Train/Test/Val detection datasets

$ cd data
$ python voc_label.py

Weights

1. yolov4-ciou.weights:https://drive.google.com/open?id=1IsLdGVkgTHl4G0EbZzb4PlIyfyyjXd9E
2. yolov4-mse.weights:https://drive.google.com/open?id=1znVte9XIyWjUGbzh1Cif8IWG4jxswn5L
3. yolov4-giou.weights:https://drive.google.com/open?id=1TTZ5Q_mJ4DYfJBp_DnoZXSMGbAJ7zIcn
4. yolov4-diou.weights:https://drive.google.com/open?id=1XEez2aGAnS60kD4VFzQ7LM8wppAdKpBj
5. (Train-from-scratch)yolov4.conv.137: https://drive.google.com/open?id=1N7_Gea1gol6ZaGfB8XBnreaeH3D9vZ6f

Train

Before you start train or test, make sure you are at the folderChris-s-darknet.

First, place model file in folder you like. Below, I place in ./weights. By running one of following lines, a loss plot will be saved in current folder with name like chart_yolov4-ciou.png:

MSE loss

$ ./build-release/darknet detector train cfg/voc.data cfg/yolov4-mse.cfg weights/yolov4.conv.137 -dont_show -map

CIOU loss

$ ./build-release/darknet detector train cfg/voc.data cfg/yolov4-ciou.cfg weights/yolov4.conv.137 -dont_show -map

IOU loss

$ ./build-release/darknet detector train cfg/voc.data cfg/yolov4-iou.cfg weights/yolov4.conv.137 -dont_show -map

Test

Here replace backup2/yolov4-ciou.weights with model you want to test.

$ ./build-release/darknet detector map cfg/voc.data cfg/yolov4-ciou.cfg backup2/yolov4-ciou.weights -dont_show -map

Test with different nms method:

If you want to test mAP with different NMS method, change three lines in which .cfg you are using:

For example, I want to test on YOLOv4 trained with CIOU loss, so I edit ./cfg/yolov4-ciou.cfg:

Optional NMS are:

1. greedynms
2. cornersnms
3. diounms

Test with different IoU threshold:

The default iou_thresh is set to 0.5. If want to test with different iou_thresh, e.g., 0.75, run:

$ ./build-release/darknet detector map cfg/voc.data cfg/yolov4-ciou.cfg backup2/yolov4-ciou.weights -dont_show -map -iou_thresh 0.75