it is very new idea in 3D reconstruction,but would u like to share me with the model apply to the real-world data,like u said in paper,and in my experience it did't work,could u share me with the details?Thanks to your reply.

val-iisc / 3d-lmnet

Repository for 3D-LMNet: Latent Embedding Matching for Accurate and Diverse 3D Point Cloud Reconstruction from a Single Image [BMVC 2018]

https://val-iisc.github.io/3d-lmnet/

MIT License

113 stars 24 forks source link

it is very new idea in 3D reconstruction,but would u like to share me with the model apply to the real-world data,like u said in paper,and in my experience it did't work,could u share me with the details?Thanks to your reply. #4

Closed AmyWan closed 5 years ago

AmyWan commented 5 years ago

I am sorry for ask again.I want to know how to use the nature images to test the model.but the demo is for shapnet.

priyankamandikal commented 5 years ago

For testing on natural images, be sure to mask out the background, then crop and resize the image to 128 x 128 dimension (keeping the aspect ratio same) before passing it through 3D-LMNet.

AmyWan commented 5 years ago

mask out

ok,I directly resize the nature image(mask out the background) to 128 *128 with function of opencv resize,is it right? I didnt do crop ,dose it have influeuce?

priyankamandikal commented 5 years ago

I have now updated the repository with the code for generating results on the real-world Pix3D dataset. Please follow the updated instructions for viewing the results. The specific pre-processing perfomed on these natural images is present here: https://github.com/val-iisc/3d-lmnet/blob/master/metrics.py#L92