shubhtuls / drc Goto Github PK
View Code? Open in Web Editor NEWCode release for "Multi-view Supervision for Single-view Reconstruction via Differentiable Ray Consistency" (CVPR 2017)
Home Page: https://shubhtuls.github.io/drc/
Code release for "Multi-view Supervision for Single-view Reconstruction via Differentiable Ray Consistency" (CVPR 2017)
Home Page: https://shubhtuls.github.io/drc/
Hi,
I have problems evaluating Pascal, seems like I missed some download. When I run the evaluation I get nil values in the matlab script (see attachment). It is also unclear how I should change the paths in https://github.com/shubhtuls/drc/blob/master/preprocess/pascal/startup.m
as in the tars there was no PASCAL3D folder I believe.
When I try to compute the ortographic cam I get:
classes = {'aeroplane','car','chair'};
for c=1:3
computeOrthographicCam(classes{c});
computeOrthographicCamImagenet(classes{c});
end
Error using load
Unable to read file '/home/meeso/drc/cachedir/Datasets/PASCAL3D/Annotations/aeroplane_pascal/2008_000021'. No such file or directory.
Error in extractPascalAnnotations (line 9)
load(fullfile(annoDir,voc_image_id));
Error in computeOrthographicCam (line 3)
pascalInstances = extractPascalAnnotations(className);
There is a dependency with libmatio that is not mentioned.
To solve it:
sudo apt-get install libmatio2
luarocks install matio
Error:
$ gpu=1 th synthetic/evalScripts.lua | bash
/home/meeso/torch/install/bin/luajit: /home/meeso/torch/install/share/lua/5.1/trepl/init.lua:389: module 'matio' not found:No LuaRocks module found for matio
no field package.preload['matio']
no file '/home/meeso/.luarocks/share/lua/5.1/matio.lua'
no file '/home/meeso/.luarocks/share/lua/5.1/matio/init.lua'
no file '/home/meeso/torch/install/share/lua/5.1/matio.lua'
no file '/home/meeso/torch/install/share/lua/5.1/matio/init.lua'
no file '/usr/share/lua/5.1/matio.lua'
no file '/usr/share/lua/5.1/matio/init.lua'
no file './matio.lua'
no file '/home/meeso/.luarocks/lib/lua/5.1/matio.so'
no file '/home/meeso/torch/install/lib/lua/5.1/matio.so'
no file '/home/meeso/torch/install/lib/matio.so'
no file '/usr/lib/lua/5.1/matio.so'
no file './matio.so'
no file '/usr/local/lib/lua/5.1/matio.so'
no file '/usr/local/lib/lua/5.1/loadall.so'
stack traceback:
Hi,
When I tried the demo.ipynb, it reports that cannot find [pascal,shapenet]_pred.png. In this code, it indeed did not save the *_pred.png . But unfortunately I'm not familiar with lua, could you please tell me how should i save this file?
And there are some more questions:
Is there any predict script which can reconstruct from a single input image ?
In your paper 2017 Hierarchical Surface Prediction for 3D Object Reconstruction, the resolution of predicted geometry increased along the increment of voxel side length. And in the script reconstructionBenchmark.lua, it has param of grid size.
So I was wondering whether the grid size will influence the resolution of predicted geometry. (Will the param of image size
also influence the resolution ?). If it's true, can i simple change this parameter to 64, 128 and so on to get a geometry with higer resolution using your provided pre-model ? (I am now struggling with downloading those two big datasets and thus cannot directly try this by changing the param in your script. So could you please give me a theoretical answer.)
In the DRC paper, you provided results of 3D Scene Recongstruction from Ego-motion in Section 5.3 and Object Reconstruction from RGB Supervision in Section 5.4.
Could you please tell me which script in this repo. can reproduce these two baseline ?
Can you please provide a working link to the pre-trained models. The provided link is not working.
When preprocessing the datasets with
classes = {'aeroplane','car','chair'}; for c=1:3 computeOrthographicCam(classes{c}); computeOrthographicCamImagenet(classes{c}); end precomputeVoxelsP3d;
one gets an error, if the Matlab Aerospace toolbox is not installed:
Undefined function or variable angle2dcm. Error in extractPascalAnnotations>viewpointToRots (line 49) R = angle2dcm(pi-euler(1), -pi/2 + euler(2), euler(3),'ZXZ'); %empirically computed Error in extractPascalAnnotations (line 26) [rot,euler]=viewpointToRots(viewpoint); Error in computeOrthographicCam (line 3) pascalInstances = extractPascalAnnotations(className);
Great work, Shubham!
When I run the codes on a single-view depth map, I test on a min-batch using only a depth map, I find that the loss can not decrease to 0, why?
When you run the preprocessing commands, you need to have the Mapping Matlab Toolbox installed, if not you encounter following error:
Undefined function or variable 'extractfield'.
Error in getFileNamesFromDirectory (line 34)
nameStruct = [nameStruct extractfield(t,'name')];
Error in computeOrthographicCam (line 20)
shapenetModelNames = getFileNamesFromDirectory(synsetDir,'types',{''});
Error in preprocess_everything (line 4)
computeOrthographicCam(classes{c});
The Image Processing Toolbox Matlab is necessary to run the preprocessing script:
Undefined function or variable 'roipoly'.
Error in computeOrthographicCam (line 47)
mask = roipoly(im, pascalInstances(ix).poly_x, pascalInstances(ix).poly_y);
Error in preprocess_everything (line 4)
computeOrthographicCam(classes{c});
Hi Dr. Shubham,
It seems the download link for segmentation masks of VOC 3D+ dataset is invalid. Could you please provide a new download link?
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.