Geo-PIFu training is too slow

simpleig / Geo-PIFu

This repository is the official implementation of Geo-PIFu: Geometry and Pixel Aligned Implicit Functions for Single-view Human Reconstruction.

111 stars 16 forks source link

Geo-PIFu training is too slow #10

Open bluestyle97 opened 3 years ago

bluestyle97 commented 3 years ago

Hi, I have finished the data preprocessing following your instruction and start to train the model. However, I found that it is too time-consuming to train Geo-PIFu. The first stage, i.e., apps.train_shape_coarse has been running for 3 days but is still at the 5th epoch (30 epochs in total). In README it is claimed that this stage only spend 2 days, so what is the problem? I train with 4 TITAN RTX GPUs and set the batch size to 16, it should not be so slow I guess.

sunjc0306 commented 3 years ago

I have the same problem. Maybe the data is too much, I only use one percent of the data.

direcf commented 3 years ago

Did you acquire satisfactory results via training and inference like paper result?

@sunjc0306 The comment 'I only use one percent of the data' means that can I get the good results from 1/30 data sampling(voxelization and mesh)? The reason why I ask you this question is that I failed to utilize conifg file because of 000072.json file. (os.path.exists doesn't figure it out)