search

microsoft / singleshotpose

This research project implements a real-time object detection and pose estimation method as described in the paper, Tekin et al. "Real-Time Seamless Single Shot 6D Object Pose Prediction", CVPR 2018. (https://arxiv.org/abs/1711.08848).
MIT License
721 stars 215 forks source link

What's the difference between LINEMOD and OCCLUSION dataset? #34

Closed kevinkit closed 6 years ago

kevinkit commented 6 years ago

The paper states that the architecture was tested on the LINEMOD and the OCCLUSION dataset. I was able to download the LINEMOD dataset, however I was unable to find OCCLUSION to see what the differences are.

It looks like they are basically the same with different annotations ? Can somebody please give me some hints for that?

Thanks in adavance

SergioRAgostinho commented 6 years ago

The Occlusion Dataset is a reannotation of the benchvise segment of the LINEMOD dataset. All objects are now annotated instead of simply the benchvise. You're likely to find more info here https://hci.iwr.uni-heidelberg.de/vislearn/iccv2015-occlusion-challenge/

Here's a direct link to the dataset, extracted from the page I previously linked.

btekin commented 6 years ago

As Sergio mentioned, the Occlusion dataset contains additional annotations for the "benchvise" partition of the LINEMOD dataset. You can find the labels of the Occlusion dataset for the objects present in the benchvise segment in LINEMOD/[OBJ_NAME]/labels_occlusion folders.

chenyu36 commented 5 years ago

I had success in creating labels (i.e. the ground truth) for custom objects, but not sure how to create the lables_occlusion similar to those in LINEMOD dataset. I am using Unity game engine to create the ground truth. @btekin Could you provide detail instructions on how to create labels for occlusion for multi obj pose estimation?

scp10086 commented 5 years ago

Help! I want to create a 3D model of my cup with intel RealSense SR300. Is there any 3D scan software (Free and useful) which support that device?

satpalsr commented 2 years ago

Hey @chenyu36, did you figure out creating labels for the occlusion dataset. I am having problem specifically with different rotation matrix in occlusion dataset for each of the object.