Comments (11)
I cannot assert whether your results are normal since I don't save the experimental results on NYUDepth V2 dataset currently. But if you have reached a RMSE around 105mm, than it seems going on well.
from penet_icra2021.
Thank you very much for your reply again! Because I watched PENet for a long time, and the teacher later asked me to use the indoor data set to complete the project. The paper of this network only used the KITTI data set for experiments, and I have many novel ideas about this network, so I especially want to to try. So I have a lot of questions, please!
I think one of the questions I want to ask the most is: Does PENet work on indoor datasets?
The second major problem: the RMSE of the current training is large. But I don't think the network can't predict the NYU depth map. I think this has something to do with the unit of NYU and KITTI depth values. I see that the depth value 'd' and 'gt' of KITTI in the code is in the style of Figure 1,2, while the 'd' in NYU is in the style of Figure 3,4
If the KITTI depth value is in m, what is the NYU depth value in? I think both m and mm will not work here, is it cm? Or I have a problem converting the NYU dataset format. Do you know what the unit of the NYU depth map is? I think the RMSE value displayed now is very large, the problem is related to the unit of the depth value of each pixel in the dataset. I don't know if what I said is clear.
Excuse me! ! !
from penet_icra2021.
I am not certain now because I didn't save the experimental records. But one thing I can ensure is that ENet works on NYU Depth V2, though we don't officially release the codes and results. And we follow similar experimental setting (like NLSPN) on the indoor dataset. You could check your settings and codes more detailly.
from penet_icra2021.
Thank you so much for confirming this to me! In this way, I think I don't have to change the network hahahaha! Thank you for your patience in answering!
I think I understand what you mean: use NLSPN to extract the NYU data set to write dataloader to obtain image data, I think that the extraction method of that network is the H5 file format.
I am now changing the format of the NYU dataset to the format of the KITTI dataset that PENet can input: I have made the following modifications, and I would like you to help me see if the changes I have made are correct, so that I can rule out the large error. s reason:
- I changed the size of the NYU dataset to 1242375๏ผthe original is 640480๏ผ
- At this time, the rgb of the NYU dataset is 24 bits, and the raw depth and gt are both 16 bits (the original is 8 bits, and the PENet format requires 16 bits of input data, so I made changes here)
- Modify the internal parameter matrix K
The above is my modification of the dataset format that PENet needs! I think you could haven't modified it this way, but I want to check with you if it's reasonable for me to do so.
I will always look forward to your reply! Thank you for not taking my trouble!
from penet_icra2021.
This is the RMSE of my first round of training ENet, which is ten times larger than the KITTI dataset, but the KITTI dataset is a few hundred for training ENet from the beginning.
from penet_icra2021.
Now I suspect that the NYU dataset is not the real depth value after changing to 16 bits, how can I get the real depth distance?
from penet_icra2021.
But I still want to ask, can you provide the code for using the NYU dataset on PENet? I will be very grateful!
from penet_icra2021.
But I still want to ask, can you provide the code for using the NYU dataset on PENet? I will be very grateful!
I conducted my experiments with the NYU Depth V2 dataloaders uploaded previously in your issues as '.txt' files. You could remove the 'txt' post-fix and add them to the dataloader directory. However we might probably not formally update codes and results on NYUDepth V2 currently.
from penet_icra2021.
Ok, thank you very much for your patient reply again!
So to say. This method of changing the format of the NYU dataset to the format of the KITTI dataset as the input to PENet does not work. I need to use other method to read NYU dataset. Is that what you mean?
from penet_icra2021.
Ok, thank you very much for your patient reply again! So to say. This method of changing the format of the NYU dataset to the format of the KITTI dataset as the input to PENet does not work. I need to use other method to read NYU dataset. Is that what you mean?
I think it is not necessary to upsample NYU figures into KITTI size. But I am not sure what point you are talking about.
from penet_icra2021.
I think I know what's wrong. I now want to experiment with KITTI first and solve that problem at the same time.
Thank you so much for your patient reply!
from penet_icra2021.
Related Issues (20)
- Projection result confirmation HOT 3
- Questions about KITTI-odometry HOT 2
- Runtime measurement HOT 4
- How to use the sparse depth?
- How to use the sparse depth? HOT 4
- Modify the backbone network HOT 3
- lightweight deployment of PENet network
- How to infer PENet for KITTI object task? HOT 3
- broken PNG file HOT 1
- the difference intrinsic parameters between train and test HOT 4
- Modify DA-CSPN++ for single branch input HOT 1
- some questions about the implement of CSPN HOT 4
- Model mismatch at inference time HOT 1
- Tensor Dimension Mismatch HOT 1
- How many parameters dose PENet have? HOT 2
- About training HOT 1
- How to implement the model with ROS?
- can't download the pretrained PENet Model
- Pretrained on NYU dataset?
- RuntimeError: CUDA out of memory.
Recommend Projects
-
React
A declarative, efficient, and flexible JavaScript library for building user interfaces.
-
Vue.js
๐ Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
-
Typescript
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
-
TensorFlow
An Open Source Machine Learning Framework for Everyone
-
Django
The Web framework for perfectionists with deadlines.
-
Laravel
A PHP framework for web artisans
-
D3
Bring data to life with SVG, Canvas and HTML. ๐๐๐
-
Recommend Topics
-
javascript
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
-
web
Some thing interesting about web. New door for the world.
-
server
A server is a program made to process requests and deliver data to clients.
-
Machine learning
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
-
Visualization
Some thing interesting about visualization, use data art
-
Game
Some thing interesting about game, make everyone happy.
Recommend Org
-
Facebook
We are working to build community through open source technology. NB: members must have two-factor auth.
-
Microsoft
Open source projects and samples from Microsoft.
-
Google
Google โค๏ธ Open Source for everyone.
-
Alibaba
Alibaba Open Source for everyone
-
D3
Data-Driven Documents codes.
-
Tencent
China tencent open source team.
from penet_icra2021.