PyTorch implementation and models for Object Pose Estimation using Mid-level Visual Representations.
Object Pose Estimation using Mid-level Visual Representations is a simpler alternative to complex pose estimation pipelines. In which the mide-level representions can be useful for different downstream tasks.
More details in the paper "Object Pose Estimation using Mid-level Visual Representations".
AVD annotation for pose and Training Instructions will be availble soon.
First stage: up-sampled and fused mid-level feature maps (not the actual normal and re-shading images) are used to predict azimuth and elevation. Second stage: The object mask and mid-level features are compared with the top three object D-masks corresponding to most likely pose hypotheses from the first stage.
The pose labelling pipeline for main object categories in AVD. Using (a) RGB and depth images of each scene, we reconstructed the (b) dense 3D point-cloud of each scene. (c) The 3D bounding boxes of objects inside these point clouds are annotated using the LabelCloud tool. (f) Poses are generated using the PNP algorithm between (d) the corners projected on the image plane and (e) corners in object coordinate frame. In total, we have labeled 6337 objects pose and 3D bounding box.
