Difficult to generate chunks

[HungNgoCT]

Hi.,

Thank you for the work.

I did experiments on my custom data, and met the problem frequently: Cannot generate chunks, although raw_chunks can.

In this experiment, I used a camera 360 and walked 3 rounds to scan. Could get good result from colmap, as shown in the image below. Then, preprocess/generate_chunks.py can generates raw_chunks, but cannot get any chunks as shown in the image. The lost of chunks also frequently occurred on my other experiments.

I think I need to adjust some parameters, but I don't know how to do.
Could you please give me some advice? and I understand scene is divided into chunks base on space information (x,y,z). Is my understanding correct? Thank you and help me to understand.

[ameuleman]

Hi,
Do you have log files log.err and log.out in raw_chunks/0_0/?
Also, the global COLMAP looks off, with a disconnected trajectory. Using the exhaustive matcher (if you are not working with too many images) or adding a few loop_matches in preprocess/make_colmap_custom_matcher.py might help.

[HungNgoCT]

Hi, Do you have log files log.err and log.out in raw_chunks/0_0/? Also, the global COLMAP looks off, with a disconnected trajectory. Using the exhaustive matcher (if you are not working with too many images) or adding a few loop_matches in preprocess/make_colmap_custom_matcher.py might help.

Hi @ameuleman .,

The colmap result is good. I walked several rounds with different paths, then choose some rounds to do this. Therefore disconnection is correct. I used camera 360 to capture. The problem here is that when it can produce raw_chunks, but cannot do chunks.

Here is 2 out of 4 folders of raw_chunks for your check if needed.
https://drive.google.com/file/d/1L3yCQkN6PBoalUdSfFKgxXv2_3inHq8f/view?usp=sharing

Or log files raw_chunks/0_0/ as you need
log.zip

[ameuleman]

Thank you. It looks like COLMAP's triangulator fails, I have not yet figured out why.

BTW, our scripts expect a folder for each camera. In your case, for each 360 panorama, we have several images (six I think?) extracted with a different angle. You should create a folder for each of these angles. I doubt it will solve the chunk processing problem, but it should improve calibration.

[HungNgoCT]

Thank you. It looks like COLMAP's triangulator fails, I have not yet figured out why.

BTW, our scripts expect a folder for each camera. In your case, for each 360 panorama, we have several images (six I think?) extracted with a different angle. You should create a folder for each of these angles. I doubt it will solve the chunk processing problem, but it should improve calibration.

Thank you @ameuleman ,

Yes, I separated into 8 images for each camera position. Will try with each images in a folder, then update you know.

[Linkersem]

Yehh, i also meet this problem.This is especially true if I have a colmap file that is computed from somewhere else, and the log.err file also shows triangulation problems.