search

uzh-rpg / rpg_open_remode

This repository contains an implementation of REMODE (REgularized MOnocular Depth Estimation), as described in the paper.
http://rpg.ifi.uzh.ch/docs/ICRA14_Pizzoli.pdf
GNU General Public License v3.0
491 stars 187 forks source link

How to process another dataset? #16

Open lucasa opened 8 years ago

lucasa commented 8 years ago

Hi. This project is very interesting. How can i use this do made depth maps to a new image sequence?

Thanks, Lucas

pizzoli commented 8 years ago

Hi Lucas, thank you for your interest. Please refer to the project wiki. If you want to process a dataset containing images and poses, then have a look at the file test/publish_dataset.cpp. You just need to adapt the input format according to your dataset syntax.

If you want to run REMODE on a live camera stream, the easiest is to

ydirey commented 8 years ago

Hi Mattia,

as a follow-up on your statement regarding running REMODE on a live camera stream. I'm trying to run your REMODE code using SVO and live camera stream.

Which tool should I use to calibrate the camera for REMODE? The calibration procedure from here http://docs.opencv.org/2.4/doc/tutorials/calib3d/camera_calibration/camera_calibration.html is not understandable for me. It presents a lot of code, but I cannot figure out how to get the calibration parameters by running it. I was able to perform the camera calibration for ATAM model using PTAM as described here: https://github.com/uzh-rpg/rpg_svo/wiki/Camera-Calibration However it provides less or different parameters then those needed in REMODE launch file.

Do you have any recommendation regarding camera settings, e.g. white-balance, focus, brightness, contrast etc?

Thank you, Yaroslav

pizzoli commented 8 years ago

Hi, REMODE needs the OpenCV camera calibration format. It is a pinhole camera model with radial/tangential distortion model. I would strongly recommend the use of a global shutter camera with fixed focus. It is definitely worth to put some effort in calibrating it properly ;) You can refer to the sample in https://github.com/Itseez/opencv/blob/master/samples/cpp/calibration.cpp, build it and running specifying a list of images. If you use ROS, then the camera calibrator is a practical tool.

Woking-34 commented 8 years ago

Hi,

I would like to setup remode to process the "Fast motion" sequence instead of the default "Over table - remode_test_data.zip" sequence. I'm using the Fast motion sequence downloaded from https://github.com/BigRedT/DTAM/tree/master/Trajectory_30_seconds. Obviously, I've altered the input parsing publish_dataset.cpp according to the new format. I've also changed the depth range. Unfortunately, I could not get any meaningful ouptut from the new setup. The camera parameters and the image size is the same as the original input.

Does anyone succeed with different dataset other than the default? Thanks!