stabilo

Stabilization settings.

Parameters for Stabilo video stabilization. Optimized for ORB + BF matching with MAGSAC++ estimation on high-altitude BEV drone imagery at ~150 m altitude. Full parameter reference: https://rfonod.github.io/stabilo/usage/#8-configuration-parameters

Generated from geotrax/cfg/default.yaml; every row below is a key of that file with its shipped default and its inline comment.

Image pre-processing

Key	Default	CLI	Description
clahe	false		Apply Contrast Limited Adaptive Histogram Equalization before feature detection; [true, false]
downsample_ratio	0.5		Resize input frames by this factor before processing; lower = faster but fewer features; (0.0, 1.0]

Feature detector and descriptor

Key	Default	CLI	Description
detector_name	'orb'	--stab-detector	Feature detector/descriptor; classical: [orb, sift, rsift, brisk, kaze, akaze], learned: [xfeat, disk, dedode, keynet, loftr]
max_features	2000		Maximum keypoints kept per frame, the strongest by response; higher = more robust but slower; [0, inf)
ref_multiplier	2.0		Multiplier applied to max_features for the reference frame only; (0.0, inf)
sift_enable_precise_upscale	false		SIFT/RSIFT only: precise sub-pixel upscaling at octave -1; [true, false]
rsift_eps	0.00000001		RSIFT only: epsilon added before RootSIFT L1-normalisation (1e-8); > 0.0, keep very small

Feature matching and filtering

Key	Default	CLI	Description
matcher_name	'bf'		Matching strategy; choices: [bf (brute-force), flann, lightglue (learned; needs disk/dedode/keynet)]
filter_type	'ratio'		Post-match filter; choices: [none, ratio (Lowe), distance]
filter_ratio	0.9		Lowe ratio-test threshold; lower = stricter (fewer but better matches); (0.0, 1.0]

Transformation estimation

Key	Default	CLI	Description
transformation_type	'projective'		Geometric model to estimate; choices: [projective, affine]. 'affine' accepts only ransac_method 4 (LMEDS) or 8 (RANSAC)
ransac_method	38		RANSAC method; [4: LMEDS, 8: RANSAC, 16: RHO, 32: USAC-Default (LO-RANSAC), 33: USAC-Parallel (LO-RANSAC), 35: USAC-Fast (LO-RANSAC), 36: USAC-Accurate (GC-RANSAC), 37: PROSAC, 38: MAGSAC++]
ransac_epipolar_threshold	2.0		Reprojection error threshold [px] for RANSAC inlier classification, in the pixels of ransac_threshold_space; (0.0, inf)
ransac_threshold_space	'full'		Pixels the threshold is measured in; ['full' (full resolution), 'processed' (downsampled image: threshold / downsample_ratio)]. They agree at downsample_ratio 1.0
ransac_max_iter	5000		Maximum RANSAC iterations; (0, inf]
ransac_confidence	0.999999		Required probability that the best model is correct; (0.0, 1.0]

Foreground mask

Key	Default	CLI	Description
mask_use	true		Exclude detected vehicle bounding boxes from feature extraction; [true, false]
mask_margin_ratio	0.15		Extra margin around each bounding box mask as a fraction of its size; [0.0, 1.0]

Detector-specific thresholds (BRISK, KAZE, AKAZE): each detects every keypoint whose response exceeds its threshold, then keeps the max_features strongest. The defaults are OpenCV's.

Key	Default	CLI	Description
brisk_threshold	30		BRISK (AGAST) detection threshold; lower = more candidate keypoints; (0, 255]
kaze_threshold	0.001		KAZE detector response threshold; lower = more candidate keypoints; (0.0, inf]
akaze_threshold	0.001		AKAZE detector response threshold; lower = more candidate keypoints; (0.0, inf]

Learning-based (kornia) detector weights; only used when detector_name is one of them. Mind the rotation and memory caveats before switching: see https://rfonod.github.io/geo-trax/guides/feature-detectors/

Key	Default	CLI	Description
loftr_weights	'outdoor'		LoFTR pretrained weights; [outdoor, indoor]
loftr_confidence	0.0		LoFTR min. correspondence confidence to keep; [0.0, 1.0] (0.0 keeps all)
disk_weights	'depth'		DISK pretrained weights; [depth, epipolar]
dedode_detector_weights	'L-C4-v2'		DeDoDe detector weights; [L-upright, L-C4, L-SO2, L-C4-v2]
dedode_descriptor_weights	'B-upright'		DeDoDe descriptor weights; [B-upright, G-upright, B-C4, B-SO2, G-C4, G-SO2]

Computation. Two mutually exclusive paths: gpu = OpenCV CUDA (classical detectors, orb only; needs a CUDA-enabled OpenCV build), device = torch device (learning-based detectors only).

Key	Default	CLI	Description
gpu	false	--stab-gpu, --no-stab-gpu	Use GPU-accelerated operations where available; [true, false]
gpu_device_id	0	--stab-gpu-device-id	CUDA device index used when gpu is true (multi-GPU machines); [0, inf)
device	'auto'	--stab-device	Torch device, learned detectors only; ['auto' (cuda > mps > cpu), 'cpu', 'cuda', 'mps']

Diagnostics

Key	Default	CLI	Description
viz	false		Retain keypoints and match visualizations in memory (for debugging); [true, false]
benchmark	false		Enable timing benchmarks per frame; [true, false]

Warning thresholds

Key	Default	CLI	Description
min_good_match_count_warning	20		Warn if filtered match count falls below this; [0, inf]
min_inliers_match_count_warning	10		Warn if RANSAC inlier count falls below this; [0, inf]