simple_visual_slam

Reference Keyframe Experiments

Generated at: 2026-04-06T12:06:09.889672+00:00

Problem

Tracking currently decides whether a newly inserted keyframe should immediately become the reference anchor. This pilot turns that decision into an experiment surface: one shared contract, one shared scenario corpus, and multiple competing policies.

Shared Inputs

Scenario corpus: experiments/reference_keyframe/scenarios.csv with 14 comparable cases.

Runtime Results

Policy Status Philosophy Accuracy Precision Recall Promote Rate Mean Confidence Mean Eval ns
heuristic core imperative-thresholds 0.857 0.818 1.000 0.786 0.794 24.47
score experiment weighted-score 0.929 1.000 0.889 0.571 0.784 29.48
pipeline experiment staged-gates 0.929 1.000 0.889 0.571 0.809 20.59

Static Proxies

Readability and extensibility are heuristic scores generated from code size, branch count, named constants, and helper-function count.

Policy Non-comment LOC Branch Points Helper Functions Named Constants Readability Extensibility
heuristic 56 2 2 2 3.59 3.04
score 86 7 7 9 1.83 5.00
pipeline 107 11 5 8 1.00 4.36

Mismatch Hotspots

Policy Mismatched Scenarios
heuristic room_mono_thin_map_support, room_depth_accel_thin_support
score room_depth_after_minor_loss
pipeline room_mono_confident_refresh

Full Replay Stability

This follow-up replays the full bounded real-trace corpus with repeat 2 and --repro-eval.

Policy Runs Mean APE Std APE
heuristic 26 0.092 0.079
score 25 0.093 0.080
pipeline 26 0.107 0.117
Mode Policy Runs Mean APE Std APE
depth heuristic 10 0.053 0.036
depth pipeline 10 0.053 0.038
depth score 9 0.056 0.033
depth_accel heuristic 6 0.058 0.039
depth_accel pipeline 6 0.058 0.041
depth_accel score 6 0.057 0.037
mono heuristic 10 0.151 0.091
mono pipeline 10 0.191 0.148
mono score 10 0.148 0.097

Room Focus Follow-Up

This hotspot follow-up replays only rgbd_dataset_freiburg1_room windows for mono and depth_accel with --repro-eval enabled.

Policy Runs Mean APE Std APE
heuristic 49 0.161 0.104
score 50 0.160 0.101
pipeline 49 0.160 0.105
Mode Policy Runs Mean APE Std APE
depth_accel heuristic 25 0.078 0.018
depth_accel pipeline 25 0.080 0.020
depth_accel score 25 0.077 0.025
mono heuristic 24 0.247 0.085
mono pipeline 24 0.243 0.092
mono score 25 0.244 0.076
Case Policy Runs Mean APE Std APE
room_depth_accel_head heuristic 5 0.102 0.012
room_depth_accel_head pipeline 5 0.104 0.012
room_depth_accel_head score 5 0.094 0.013
room_depth_accel_late heuristic 5 0.056 0.009
room_depth_accel_late pipeline 5 0.055 0.007
room_depth_accel_late score 5 0.061 0.005
room_depth_accel_mid heuristic 5 0.087 0.008
room_depth_accel_mid pipeline 5 0.075 0.004
room_depth_accel_mid score 5 0.074 0.007
room_depth_accel_recovery heuristic 5 0.078 0.004
room_depth_accel_recovery pipeline 5 0.081 0.007
room_depth_accel_recovery score 5 0.077 0.006
room_depth_accel_tail heuristic 5 0.068 0.004
room_depth_accel_tail pipeline 5 0.084 0.020
room_depth_accel_tail score 5 0.079 0.047
room_mono_head heuristic 5 0.358 0.072
room_mono_head pipeline 5 0.352 0.107
room_mono_head score 5 0.316 0.063
room_mono_late heuristic 5 0.147 0.028
room_mono_late pipeline 5 0.147 0.036
room_mono_late score 5 0.139 0.012
room_mono_mid heuristic 5 0.213 0.019
room_mono_mid pipeline 4 0.225 0.010
room_mono_mid score 5 0.229 0.028
room_mono_recovery heuristic 5 0.267 0.054
room_mono_recovery pipeline 5 0.242 0.067
room_mono_recovery score 5 0.279 0.047
room_mono_tail heuristic 4 0.251 0.040
room_mono_tail pipeline 5 0.246 0.041
room_mono_tail score 5 0.256 0.065
This site is open source. Improve this page.