This is an implementation of Deep-6D pose on Python 3, Keras, and TensorFlow. The model generates bounding boxes, segmentation masks and 6 DoF pose for each instance of an object in the image. It's based on Feature Pyramid Network (FPN) and a ResNet101 backbone.
This is not a official implementation, and I forked and modified code from Mask R-CNN [https://github.com/matterport/Mask_RCNN]
I trained and tested with YCB Object Dataset [http://www.ycbbenchmarks.com/] To generate the dataset, run the script below.
cd samples/ycb
python3 generateData.py
To train the model, run the script below.
cd samples/ycb
python3 ycb.py train --dataset=/path/to/ycb/dataset --weights=coco
To test the model, run the script below.
cd samples/ycb
python3 ycb.py mask --dataset=/path/to/ycb/dataset --weights=last
I did not implement the evaluation code.