jhornauer / grumodepth Goto Github PK
View Code? Open in Web Editor NEWGradient-based Uncertainty for Monocular Depth Estimation (ECCV 2022)
License: Apache License 2.0
Gradient-based Uncertainty for Monocular Depth Estimation (ECCV 2022)
License: Apache License 2.0
Hi @jhornauer
I am facing an issue when running the generate_maps.py
In line 487, when calling the get_mono_ratio function we are trying to resize the reciprocal of pred_disps to the gt_depths. But the cv2.resize function as defined is not working and throwing an error saying : IndexError: too many indices for array: array is 2-dimensional, but 3 were indexed
In my opinion I found this to be the case because the resize parameters as defined in the function (get_mono_ratio) are not giving the right image dimensions.
Can you please help with the issue.
Thank you,
Shubham
Hi @jhornauer
I had doubts in my understanding regrading what AURG means and what it signifies.
First the paper says that AURG is an estimation if we assume there is no modelling in the system. Here no modelling means, not arranging the uncertainty values in descending order or something else? what does it mean intuitively when we say that higher the value of AURG the better?
Thank you for your time,
Shubham
Hello,
I'm exploring depth estimation models and came across the Dense Prediction Transformer (DPT) architecture used in Omnidata's depth estimation tasks. I'm interested in enhancing the robustness of depth predictions through uncertainty estimation.
Given the gradient-based uncertainty estimation capabilities of GrUMoDepth, I am curious about its compatibility with DPT models. Specifically, I would like to know:
I appreciate any insights or recommendations on integrating these methodologies to improve the reliability and interpretability of depth estimations in complex environments.
Thank you!
Why does depth estimation layer doesn't have a sigmoid?
Do you have sigmoid after uncertconv
layer?
Shouldn't be there a relu activation after the uncertainty estimation in the supervised scenario. Since the network could predict negative uncertainty which is not intuitive; and would result in negative loss. Or is this intended behavior ?
Thank you for sharing the codes and the models.
While reading the code I have a remark regarding the python file evaluate_supervised.py. Is there a specific reason you are testing separately if opt.grad in lines 204 and 217?
Thanks.
Dear Authors, @jhornauer
Thank you for the insightful paper. I was trying to implement this on a different dataset. In the process of creating a virtual Conda env, when I am trying to install packages from the .yml file around 50% are saying no version available.
Do you know what the reason could be for this?
Thank you,
Shubham
Thank you for sharing the codes. BTW the paper is clear and intuitive.
However, I did not find the results for MS (monocular + stereo supervision) on KITTI dataset in the paper, while in mono-uncertainty paper, the authors provided them. I know this is an additional request, but since usually MS has better depth accuracy than M and S, maybe you can provide its uncertainty results in the supplementary material?
Thank you.
A declarative, efficient, and flexible JavaScript library for building user interfaces.
๐ Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
An Open Source Machine Learning Framework for Everyone
The Web framework for perfectionists with deadlines.
A PHP framework for web artisans
Bring data to life with SVG, Canvas and HTML. ๐๐๐
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
Some thing interesting about web. New door for the world.
A server is a program made to process requests and deliver data to clients.
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
Some thing interesting about visualization, use data art
Some thing interesting about game, make everyone happy.
We are working to build community through open source technology. NB: members must have two-factor auth.
Open source projects and samples from Microsoft.
Google โค๏ธ Open Source for everyone.
Alibaba Open Source for everyone
Data-Driven Documents codes.
China tencent open source team.