drsleep / multi-task-refinenet Goto Github PK

I want to know how to deal the train process？And how to make the multi-task loss?
Can you give us some help?

I was wondering if the output is in meter or I have to convert it. Thank you for sharing yout project!

Hi,

Thank you for sharing codes!

I am wondering how you conducted experiments and how you get the results of previous methods in Table 1.
Did you use official codes or did you reimplement previous methods by yourself?

Hello
I was wondering that how did you calculate the depth_scale = 5000 and how did you calculate max and min depth.

Hi, I transform the jupyter notebook to python code. And I run the project.
However, I got the segmentation result which has the different color from yours. Do you know why?

Could you offer a train.py? thank you very much!

Any updates on the Train process ? or any guidance is highly appreciated .

Could you please release the pretrained models, especially the "Single Model - Two Datasets, Two Tasks"? Thanks a lot.

There are 464 different scene types in NYUv2 raw dataset, where 249 scenes are for training and 215 scenes are for validation.
Common implementation is to use only the images from 249 scenes during the training process.
I get very high mIoU and PA with your provided model, so I wonder whether you include the 215 validation scenes in your pre-training procedure on the large dataset.

I dont know how to change https://github.com/DrSleep/light-weight-refinenet/blob/master/src/train.py to run https://github.com/DrSleep/multi-task-refinenet/blob/master/src/models.py (Especially the part of dataloader)

would you like to share the train process?
It would be better to have a valid process.

Hi I wanted to know about your dataloader part for NYUD dataset
I am Normalizing and converting to tensor, in my ToTensor() function i am reading the data and converting them to float.

is this correct, if possible please the dataloaders used in this papers