This project aims to develope a detector cabable of detecting and predicting the 6D pose of objects in RGB images. The structur of the detector will be based on the RetinaNet detector proposed in Focal Loss for Dense Object Detection. As a starting point the Keras RetinaNet implementation provided by Fizyr will be used.
- Clone this repository.
- Ensure numpy is installed using
pip install numpy --user - In the repository, execute
pip install . --user. Note that due to inconsistencies with howtensorflowshould be installed, this package does not define a dependency ontensorflowas it will try to install that (which at least on Arch Linux results in an incorrect installation). Please make suretensorflowis installed as per your systems requirements. - Alternatively, you can run the code directly from the cloned repository, however you need to run
python setup.py build_ext --inplaceto compile Cython code first.
The detector will be benchmarked on the LINEMOD dataset.
The developed detectors performance on the LINEMOD dataset will be compared to the results introduced in the acrticle Real-Time Seamless Single Shot 6D Object Pose Prediction which is considered to be state of the art.
The CSVGenerator provides an easy way to define your own datasets.
It uses two CSV files: one file containing annotations and one file containing a class name to ID mapping, path to point_cloud_file and the length of the diagonal of the point cloud.
path/to/image.jpg,x1,y1,x2,y2,class_name, R00, R01, R02, R10, R11, R12, R20, R21, R22, T0, T1, T2
(x1, y1), (x2, y2) - Represents upper left and lower right image coordinates of the 2D bounding box enclosing the object
(R00, R01, R02); (R10, R11, R12); (R20 R21, R22)- Represents the rotations matrix
(T0; T1; T2) - Represents the translations vector
The class name to ID mapping file should contain one mapping per line. Each line should use the following format:
class_name,id,path/to/point_cloud.npy,diag_distance