ElonZhou

官方测试的精度是量化之后的结果吗

7

请问在文章中提供的精度数值是量化之后的计算结果还是未量化的计算结果？

why the input of the build_targets2(aux_head) is the same as the build_target(lead_head)?

3

in the paper, loss of the aux_head is calculated by the pre of lead_head and aux_head. but in the loss.py, the build_targets2 and build_targets have the same input: pre and...

What are the diferences between train.py and train_aux.py?

1

What are the diferences between train.py and train_aux.py?

Why is the reproduction accuracy not as good as in the paper?

10

Why is the reproduction accuracy not as good as in the paper?

请问可以教一下如何处理数据集吗？谢谢。下载 data_object_image_3.zip 是吗

how to treat video datasets differently with the same as paper?

3

Excuse me. In the DATA.MD, there is the adress to download the video files. But do not tell us how to treat video datasets differently. Or can you provide me...

How to set the config to use multi node and multi GPUS?

1

Hi, How to set the training config to use multi node and multi GPUS?

https://github.com/cliport/cliport/blob/2be5c47b5bb9bb7040ad90693288b87b1e18e7ad/cliport/tasks/task.py#L132 We visualised the result of this `pick_pix = utils.sample_distribution(pick_prob)` calculation of pick_pix in the front color image and found that the position is not on the grabbed object, may...

About the dataset split of the grasp anything++ training and val.

9

Glad you shared such a great job. I noticed in the code that the number of `split/grasp-anything++/train/seen.obj` is 14516, the number of `split/grasp-anything++/test/seen.obj` is 573, and the number of `split/grasp-anything++/test/unseen.obj`...