How to run with own data?

[ghost]

Hi and thank you for the release,
I succesfully installed it and prepared a dataset with data/create_data_own.py
When I run the pretrained network as python3 exp.py --net resunet3.16_penone.dirs.avg.seq+9+1+unet+5+2+16.single+mlpdir+mean+3+64+16 --cmd eval --iter last --eval-dsets mydataset
The script silently exits after these messages

[2021-04-22/15:51/INFO/modules] [NET][EncNet] resunet3.16
[2021-04-22/15:51/INFO/modules] [NET][RefNet] point_edges_mode=penone
[2021-04-22/15:51/INFO/modules] [NET][RefNet] point_aux_data=dirs
[2021-04-22/15:51/INFO/modules] [NET][RefNet] point_avg_mode=avg
[2021-04-22/15:51/INFO/modules] [NET][RefNet] Seq 9 nets, nets_residual=True
[2021-04-22/15:51/INFO/modules] [NET][RefNet]   Unet(in_channels=16, enc_channels=[16, 32, 64, 128, 128], dec_channels=[128, 64, 32, 16], n_conv=2)
[2021-04-22/15:51/INFO/modules] [NET][RefNet]   Unet(in_channels=16, enc_channels=[16, 32, 64, 128, 128], dec_channels=[128, 64, 32, 16], n_conv=2)
[2021-04-22/15:51/INFO/modules] [NET][RefNet]   Unet(in_channels=16, enc_channels=[16, 32, 64, 128, 128], dec_channels=[128, 64, 32, 16], n_conv=2)
[2021-04-22/15:51/INFO/modules] [NET][RefNet]   Unet(in_channels=16, enc_channels=[16, 32, 64, 128, 128], dec_channels=[128, 64, 32, 16], n_conv=2)
[2021-04-22/15:51/INFO/modules] [NET][RefNet]   Unet(in_channels=16, enc_channels=[16, 32, 64, 128, 128], dec_channels=[128, 64, 32, 16], n_conv=2)
[2021-04-22/15:51/INFO/modules] [NET][RefNet]   Unet(in_channels=16, enc_channels=[16, 32, 64, 128, 128], dec_channels=[128, 64, 32, 16], n_conv=2)
[2021-04-22/15:51/INFO/modules] [NET][RefNet]   Unet(in_channels=16, enc_channels=[16, 32, 64, 128, 128], dec_channels=[128, 64, 32, 16], n_conv=2)
[2021-04-22/15:51/INFO/modules] [NET][RefNet]   Unet(in_channels=16, enc_channels=[16, 32, 64, 128, 128], dec_channels=[128, 64, 32, 16], n_conv=2)
[2021-04-22/15:51/INFO/modules] [NET][RefNet]   Unet(in_channels=16, enc_channels=[16, 32, 64, 128, 128], dec_channels=[128, 64, 32, 16], n_conv=2)
[2021-04-22/15:51/INFO/modules] [NET][RefNet] Single gnn
[2021-04-22/15:51/INFO/modules] [NET][RefNet]   MLPDir(in_channels=16, hidden_channels=64, n_mods=3, out_channels=16, aggr=mean)
[2021-04-22/15:51/INFO/modules] [NET][RefNet] out_conv(16, 3)
[2021-04-22/15:51/INFO/mytorch] [EVAL] loading net for iter last: experiments/tat-wo-val_bs1_nbs3_rpointdir_s0.25_resunet3.16_penone.dirs.avg.seq+9+1+unet+5+2+16.single+mlpdir+mean+3+64+16_vgg/net_0000000000000000.params

Seems like it doesnt find my dataset. Any advice?

[ptc-lexvandersluijs]

Hi @vuoriov4, great to see that you were able to prepare your own dataset!

From my side, I have to admit that I'm a bit of a n00b what Python is concerned. Would you be willing to write a short description of what you did to go from a COLMAP result to NVS images using SVS? I would appreciate it greatly!

[ghost]

Hi @vuoriov4, great to see that you were able to prepare your own dataset!

From my side, I have to admit that I'm a bit of a n00b what Python is concerned. Would you be willing to write a short description of what you did to go from a COLMAP result to NVS images using SVS? I would appreciate it greatly!

Hey. Have you looked at data/create_data_own.py? I had to modify alot of things though, because they are hardcoded to use the benchmark datasets. Overall this repo is not very friendly for custom datasets. Not sure why you are interested in SVS, you may want to look at IBR-Net which is more recent and better documentation.

[ptc-lexvandersluijs]

Well, an important reason is that SVS should be very fast: 1 second to generate a novel view, and (like IBRNet) no training time for individual datasets, just some pre-processing of the images that only needs to be done once.
From what I read, IBRNet takes somewhere from 30 seconds to a minute to generate an image. Whereas with a runtime of one second, I may be able to apply the technology in a practical use-case.

[alex04072000]

Hi @vuoriov4 and @ptc-lexvandersluijs
Did you solve the issue?
I am facing the same problem.
There is no script for generating delaunay_photometric.ply in create_data_own.py.
It seems the data preparation process is still a mystery...