Hello. I am a student studying Hand-Object mesh.

First of all, congratulations on your research success.

I have a few questions and I would appreciate it if you could answer them.

1. When will the learning code be released?
2. It seems that Penetration, Vertex, Translation, Pose, and KLD loss are used when learning CVAE in the paper. Could you please tell me what the weight was?

We will wait for your reply. thank you.

Hi, thanks for the great work! I'm reproducing the numbers of HO-3D in Table 1. And I have several questions.

1. For each of the ten objects, how many samples should I use during evaluation? 100000?
2. Am I correct to re-initialize the cVAE decoder weights at the begining of each sequence(object) and keep it being optimized over the whole sequence?
3. Will the batch-size, the order of samples and the augmentation of samples affect the evaluation results?

Thanks!

I am trying to visual the diverse grasp. However, the visualized policy seems wrong for some reason, I just follow the README, have no idea why this happen. I would be appreciate if some can help!

Hi, I think the metric of intersection volume is not correct. The function intersect_vox is supposed to compute the intersection volume. However, vox=mesh.voxelized() and vox.points are only giving the surface points. So it is not offering a solid volume. The mesh.contains(points) is giving a smaller estimate of intersection volume.

https://github.com/hwjiang1510/GraspTTA/blob/cecb9642e6d63670d4e954cf420d03f1a93b5a90/metric/intersect.py#L7-L12

你好， 最近关注到您的文章——Grasp TTA，这对我们目前的研究工作有较强的参考价值，因此希望能够复现其效果，但该工作还没有release模型训练代码，希望能获得该部分代码，我们将在未来的文章中引用您的paper。 感谢！

Hello I recently noticed your work "Grasp TTA", which has strong reference value for our current research work. Therefore, I hope to reproduce its. However, this work has not released the model training code, we hope to get this part of the code. We will refer to your paper in future articles. Thank you!

I admire you very much for this research work, but I encountered some problems in evaluating the Obmanc test set.

When I evaluated the GT of Obman test, I found that the Penetration Volume calculated by my side does not correspond to your article, the value in your article is 1.70cm³, and the value calculated by my side is 2.68cm³.

I would like to confirm the details with you, Obman test set a total of 6285 samples, each sample has Hand and Object one-to-one correspondence, I used the intersect_vox function in your evaluation code to evaluate the GT, please ask me whether my side ignored something, which led to the calculated value is different.

Hello, firstly I want to mention this is some great work!

I was using the trained model to generate grasps for my own object pointclouds generated from simulation. Surprisingly the generated hand vertices were very distant from the object. I am not very sure the reason for this, is there any requirement on the input object pointcloud origin and axes orientation before using the network that I may have missed?

I am attaching an image of the predicted grasp for one of the input object pointcloud I used: