6dposeDetect

A 6D-pose estimation method based on fast-rcnn.

demo video

youtube:https://youtu.be/AyveVaFebcs youku:https://v.youku.com/v_show/id_XNDcyODkzMzk4NA==.html

Test method：

1.Download code from github

2.Install environment through pip

pip install -r requirements.txt

3.Download trained weights

GOOGLE DRIVE put them in: 6dposeDetect/output/res101/voc_2007_trainval/default

4.Download data

GOOGLE DRIVE put them in: 6dposeDetect/data/

5.

cd tools
python linemodocclusion9.py

You can see the visual results in the folder ./tools/show

Train method：

Since there are only 1214 images in line-occulution, we need to use data enhancement to increase the training map. I have one that has been processed. You can also try it yourself according to the project.

1.Download data

put them in： BAIDU DRIVE，key:l8j3 6dposeDetect/data/VOCdevkit2007/VOC2007 ** The data package is very large. You can decompress it like this：

cat test.tar.gz* | tar -xzv

2.

./experiments/scripts/train_faster_rcnn.sh 0 pascal_voc res101