search

fdp0525 / G2LTex

Code for CVPR 2018 paper --- Texture Mapping for 3D Reconstruction with RGB-D Sensor
146 stars 30 forks source link

Invalid CAM File Error #1

Open al093 opened 5 years ago

al093 commented 5 years ago

Hi, I am trying to run the executable exactly as written in the Readme, but i get invalid CAM file error as follows:

$ ./G2LTex ../Data/bloster/textureimages ../Data/bloster/bloster.ply tok:../Data/bloster/textureimages tok:../Data/bloster/bloster.ply PLY Loader: comment VCGLIB generated Reading PLY: 6972 verts... 12712 faces... done. Generating texture views: Invalid CAM file: color_01.cam

fdp0525 commented 5 years ago

@al093 I have fixed this mistake. Thanks.

al093 commented 5 years ago

@fdp0525 I tried to run it again, but still I get the same error. Then I checked the .cam files of the bloster dataset. It seems to miss the intrinsic matrix information, which mvs-texturing expects. Hence, I edited them and I finally could run your code.

But, I have another issue now. I am trying to run an another, my own dataset, but the tool crashes without any useful log. See the output below:

Reading PLY: 122654 verts... 242149 faces... done. Generating texture views: depth fx:934.034 fy:934.034 cx:635.012 cy:362.089 width:1280 height:720 RGB fx:1383.45 fy:1382.48 cx:979.023 cy:529.022 width:1920 height:1080 Global depth:100 RGB:1 Local RGB:1 edge:1 Reg:100 Global Iterations:300 Local Iterations:20 num_views:5 361543 total edges. 132187 faces have not been seen by a view. Saving model... done. loading face info ... Global iter:0/300 Segmentation fault (core dumped)

I dont know if its a issue with the way the .cam files must be written. Because for the translation vector in the extrinsic pose of the views I saw that you have quite small values in the bloster dataset: -2.17637 -1.55773 -0.15426 .....; are they supposed to be in meters? I cant make any sense out of these values.

But since mvs-texturing works fine on this dataset, i expect this code to run also.

fdp0525 commented 5 years ago

@al093 I have modified this issue. The extrinsic poses are in meters which are similar to the trajectory generated by KinectFusion. Thanks.

mohamed-elkhouly commented 5 years ago

Dear @fdp0525 I had the same problem of @al093 of segmentation fault, I figured that this was due to the documentation on readme: the depth naming convention was written as "Depth frames (depth-XX.png)" with "-" but your code search for depth files named with "_". So, Kindly change it in the documentation.

fdp0525 commented 5 years ago

@mohamed-elkhouly I have corrected this mistake. thanks!