svo2 parameters for higher resolution

[alex555176]

Dear all, I am working on svo2, but i met some problems on modifying parameters for different resolution of my pinhole monocular camera. Since camera resolution is 720(hight) by 1280(width) and here is parameters on the below. It will be very appreciate if anyone can tell me where is the problem or somewhere I need to modify.

\\\\\\\ pinhole.yaml ///////////////////////// //# Instructions: //# - Most of the time you will just need to change the basic parameters. //# - The parameters in this file are set for the resolution of 752x480. //# If you use a different resolution, change the parameters for //# each module according to the comments.

//############################ //##### Basic parameters ##### //############################

//# Pipeline type pipeline_is_stereo: False

//# Feature and keyframe number //# To run faster, you can decrease max_fts and max_n_kfs, for example: //# max_fts: 120 //# max_n_kfs: 5 max_fts: 120 max_n_kfs: 5

//# Map scale when initialized (not used for stereo) //# Increase if the initial scene depth is larger map_scale: 1.0

//# Initial rotation T_world_imuinit/qx: 0 T_world_imuinit/qy: 0 T_world_imuinit/qz: 0 T_world_imuinit/qw: 1

//# Keyframe selection kfselect_criterion: FORWARD # alterntive: DOWNLOOKING //# The following kfselect_* ONLY affects FORWARD //# If the number of features: >upper, no keyframe; <lower, take keyframe kfselect_numkfs_upper_thresh: 120 kfselect_numkfs_lower_thresh: 70 //# If the current frame is within the following range of any visible keyframes, //# do not take a new keyframe. //# You can lower these values for more keyframes kfselect_min_dist_metric: 0.1 kfselect_min_angle: 20 //# If the median disparity from the last keyframe is smaller, //# do not take a new keyframe //# You can lower this value for more keyframes kfselect_min_disparity: 40 //# When taking a new keyframe, //# we can already update the newly initialized seeds with old keyframes. //# This is important for forward looking case. update_seeds_with_old_keyframes: True

//# Default affine compensation parameters //# Most of the parameters are set to false for the consideration of processing time. //# If you observe bad tracking because of light/expousre change, //# enable the following parameters. img_align_est_illumination_gain: false img_align_est_illumination_offset: false depth_filter_affine_est_offset: true depth_filter_affine_est_gain: false reprojector_affine_est_offset: true reprojector_affine_est_gain: false

########################################

Tuning for each module

######################################## //# Initialization //# For a higher resolution, increase the value accordingly. init_min_disparity: 50 #30

//# Feature Detection grid_size: 30 # Larger for larger images, for every cell you have max one feature. n_pyr_levels: 4 #3 # Increase for larger images (image align max minus one) detector_threshold_primary: 10 # Fast detector threshold detector_threshold_secondary: 200 # Edgelet detector threshold

//# Image Alignment img_align_max_level: 5 # Increase this level by one if the image is double the width and height (752x480).1 img_align_min_level: 2

//# Reprojection use_async_reprojectors: False # For stereo, change it to True for multithreading. reprojector_max_n_kfs: 5 # Local map size. Larger is computationally more intensive. More reduces drift. scan_epi_unit_sphere: False # for wide angle lens, set it to True.

//# Pose Optimization poseoptim_thresh: 2.0 # Reprojection outlier threshold (px), should be larger for larger images. poseoptim_using_unit_sphere: False # For wide angle lens, set it to True

//# Depth Filter //# How many times does the covariance need to decrease until a seed is considered converged. //# Increase to get more accurate points seed_convergence_sigma2_thresh: 200

//# IMU //# Normally the prior should be 0 if not using IMU. use_imu: False poseoptim_prior_lambda: 0.0 # Gyroscope prior in pose optimization img_align_prior_lambda_rot: 0.0 # Gyroscope prior in sparse image alignment img_align_prior_lambda_trans: 0.0 # Constant velocity assumption prior in sparse image alignment

//# Dense input publish_every_nth_dense_input: 5

//# Visualization publish_marker_scale: 0.5

//# Stereo parameters, not used for mono. max_depth_inv: 0.05 min_depth_inv: 1.0 mean_depth_inv: 0.3