2
$\begingroup$

In the case of single shot detection of point clouds, that is the point cloud of an object is taken only from one camera view without any registration. Can a Convolutional Network estimate the 6d pose of objects (initially primitive 3D objects -- cylinders, spheres, cuboids)?

The dataset will be generated by simulating a depth sensor using a physics engine (ex:gazebo) and primitive 3D objects are spawned with known 6d pose as ground truth. The resulting training data will be the single viewed point cloud of the object with the ground truth label (6d pose)?

$\endgroup$
0
$\begingroup$

The answer is yes this is possible and here are the papers where they do almost exactly the same project you are describing above. Although none of the bellow combine gazebo, single point/single shot, 6D-pose and CNNs. In order to use synthetic data to train a model that works on real data.

The model will be able to be trained but how effect a model trained on the synthetic data will be able to properly function on real data will be the challenge.

$\endgroup$
  • 1
    $\begingroup$ Thanks for your reply, imo synthetic data is easier to produce, have accurate ground truth and can be manipulated by adding noise or using some smoothing to simulate real data. will definitely give it a try. Thanks again $\endgroup$ – Mostafa Said Nov 13 '19 at 20:07
  • $\begingroup$ I definite agree with you I would just be skeptical of the effectiveness of the model in specific environments, textured shadow, detecting semi translucent objects etc, normal image detection difficulties. $\endgroup$ – Michael Hearn Nov 13 '19 at 22:52

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.