taskonomy datasets does not contain mask_valid task.

EPFL-VILAB / omnidata

A Scalable Pipeline for Making Steerable Multi-Task Mid-Level Vision Datasets from 3D Scans [ICCV 2021]

Other

418 stars 53 forks source link

taskonomy datasets does not contain mask_valid task. #4

Closed FYtrace closed 2 years ago

FYtrace commented 2 years ago

Thanks for your excellent work. I met some issues when running your code. Details as follows:

First. taskonomy datasets does not contain mask_valid task. https://github.com/EPFL-VILAB/omnidata-tools/blob/7b77a4fba2fbf55b22ed4af226125d5255e55330/omnidata_tools/torch/train_depth.py#L107

https://github.com/EPFL-VILAB/omnidata-tools/blob/7b77a4fba2fbf55b22ed4af226125d5255e55330/omnidata_tools/torch/train_depth.py#L348

could I generate mask_valid according to gt_depth? sets 2**16-1 as invalid pixel?

Second. pytorch_lightning throws validation_step() missing 1 required positional argument: 'dataset_idx' error. https://github.com/EPFL-VILAB/omnidata-tools/blob/7b77a4fba2fbf55b22ed4af226125d5255e55330/omnidata_tools/torch/train_depth.py#L192

Looking forward to your reply~

Ainaz99 commented 2 years ago

Hi @FYtrace !

Yes, the invalid pixels have maximum depth value (2**16 - 1) and you can generate the masks just like you said. And for your second question, I think it's a version issue. What pytorch_lightning version are you using?

FYtrace commented 2 years ago

Hi， @Ainaz99 , Thanks for your quick reply !

My pytorch_lightning version is 1.1.8. I also tried some other latest versions, but all failed. Now, I successfully ran by deleting argument 'dataset_idx'. Maybe I should merge the val sets into single one.

Thanks for your wonderful work again. I will continue to follow your work and do something interesting.

Ainaz99 commented 2 years ago

Yes, merging the validation sets can be a solution!

icoz69 commented 2 years ago

hi, how do you handle the second issue? i came across the same problem,

puyiwen commented 2 years ago

感谢您的出色工作。我在运行您的代码时遇到了一些问题。详情如下：

第一的。taskonomy 数据集不包含 mask_valid 任务。 https://github.com/EPFL-VILAB/omnidata-tools/blob/7b77a4fba2fbf55b22ed4af226125d5255e55330/omnidata_tools/torch/train_depth.py#L107 https://github.com/EPFL-VILAB/omnidata-tools/blob/7b77a4fba2fbf55b22ed4af226125d5255e55330/omnidata_tools/torch/train_depth.py#L348

我可以根据 gt_depth 生成 mask_valid 吗？将 2**16-1 设置为无效像素？

第二。pytorch_lightning 抛出validation_step() missing 1 required positional argument: 'dataset_idx' 错误。 https://github.com/EPFL-VILAB/omnidata-tools/blob/7b77a4fba2fbf55b22ed4af226125d5255e55330/omnidata_tools/torch/train_depth.py#L192

期待您的回复~

Hi,how can I set the depth value (216 - 1) as invalid pixels?Like set the depth value as 0 which is (216 -1)?

wtpro commented 10 months ago

Hi， @Ainaz99 , Thanks for your quick reply !

My pytorch_lightning version is 1.1.8. I also tried some other latest versions, but all failed. Now, I successfully ran by deleting argument 'dataset_idx'. Maybe I should merge the val sets into single one.

Thanks for your wonderful work again. I will continue to follow your work and do something interesting.

Hi, can you elaborate on how to solve the dataset_idx issue? Simply by deleting the argument did not solve the issue at all because validation_step returns a variable that depends on it.