/object-detection-tutorial

Primary LanguagePython

object-detection-tutorial

This is a simple object detection tutorial. In this tutorial, you will learn some basic knowledge and technology about object detection. In addition, I have achieved a few simple OD models whose style are close to YOLO(You Only Look Once).

For researchers in China, you can learn this tutorial with my column in zhihu:

https://zhuanlan.zhihu.com/p/91843052

Installation

  • My PyTorch version is 1.1.0.
  • My anaconda3 version is 5.1.0 whose python is 3.6
  • You'd better to install tensorboard > 1.13.
  • For testing, I use opencv in python to visualize the detection results. So, you also need to install python-opencv.

Dataset

Because this is just a tutorial, I really do not want to download MSCOCO datatset. Just PASCAL VOC2007 and 2012.

VOC Dataset

I copy the download files from the following excellent project: https://github.com/amdegroot/ssd.pytorch

Download VOC2007 trainval & test

# specify a directory for dataset to be downloaded into, else default is ~/data/
sh data/scripts/VOC2007.sh # <directory>

Download VOC2012 trainval

# specify a directory for dataset to be downloaded into, else default is ~/data/
sh data/scripts/VOC2012.sh # <directory>

Train

  • For example, if you want to train yolo-v1 designed by myself:
python train_voc.py

if yolo-anchor:

python train_voc.py -v yolo_anchor

The default backbone network is resnet18. If you want to use darknet19:

python train_voc.py -bk d19

The model will be saved in weights/ by default.

Test

For example, you want to test the yolo-v1 model on VOC2007 test:

python test_voc.py --trained_model [ Please write down your trained model dir. ]

Evaluation

For example, you want to evaluate the yolo-v1 model on VOC2007 test:

python eval_voc.py --trained_model [ Please write down your trained model dir. ]

Train your own dataset

First, you need to make a VOC-style dataset. The names of your images are as following( .png or .jpg or whichever image format you like ):

000000.png, 000001.png, 000002.png, ...

And your annotations' names are as:

000000.xml, 000001.xml, 000002.xml, ...

Then, you need to write your dataloader file. You can modify data/voc0712.py to fit your dataset. What you need to do is just to change the path to images and annotations.

Finally, please modify the train_voc.py.