Moved 2D confidence scaling to a later step.

This commit is contained in:
Daniel
2025-03-24 14:16:08 +01:00
parent 7e3151530e
commit 0841713e1a
3 changed files with 32 additions and 16 deletions

View File

@ -828,6 +828,27 @@ std::vector<std::vector<std::array<float, 4>>> TriangulatorInternal::triangulate
auto [pose3d, score] = triangulate_and_score(
pose1, pose2, cam1, cam2, roomparams, {});
// Scale scores with 2D confidences
// They can improve the merge weighting, but especially the earlier step of pair-filtering
// works better if only per-view consistency is used, so they are not included before.
for (size_t j = 0; j < pose3d.size(); ++j)
{
float score1 = pose1[j][2];
float score2 = pose2[j][2];
float min_score = 0.1;
if (score1 > min_score && score2 > min_score)
{
float scoreP = (score1 + score2) * 0.5;
float scoreT = pose3d[j][3];
// Since the triangulation score is less sensitive and generally higher,
// weight it stronger to balance the two scores.
pose3d[j][3] = 0.9 * scoreT + 0.1 * scoreP;
}
}
all_full_poses[i] = std::move(pose3d);
}
@ -1430,11 +1451,7 @@ std::pair<std::vector<std::array<float, 4>>, float> TriangulatorInternal::triang
if (mask[i])
{
float scoreT = 0.5 * (score1[i] + score2[i]);
float scoreP = 0.5 * (pose1[i][2] + pose2[i][2]);
// Since the triangulation score is less sensitive and generally higher,
// weight it stronger to balance the two scores.
pose3d[i][3] = 0.9 * scoreT + 0.1 * scoreP;
pose3d[i][3] = scoreT;
}
}

View File

@ -41,7 +41,7 @@ default_min_bbox_score = 0.3
# Describes how good two 2D poses need to match each other to create a valid triangulation
# If the quality of the 2D detections is poor, use a lower value
default_min_match_score = 0.91
default_min_match_score = 0.94
# Describes the minimum number of camera pairs that need to detect the same person
# If the number of cameras is high, and the views are not occluded, use a higher value
@ -55,7 +55,7 @@ datasets = {
"human36m": {
"path": "/datasets/human36m/skelda/pose_test.json",
"take_interval": 5,
"min_match_score": 0.92,
"min_match_score": 0.95,
"min_group_size": 1,
"min_bbox_score": 0.4,
"min_bbox_area": 0.1 * 0.1,
@ -68,7 +68,7 @@ datasets = {
# "cams": ["00_03", "00_06", "00_12", "00_13", "00_23", "00_15", "00_10", "00_21", "00_09", "00_01"],
# "cams": [],
"take_interval": 3,
"min_match_score": 0.92,
"min_match_score": 0.95,
"use_scenes": ["160906_pizza1", "160422_haggling1", "160906_ian5"],
"min_group_size": 1,
# "min_group_size": 4,
@ -79,25 +79,25 @@ datasets = {
"path": "/datasets/mvor/skelda/all.json",
"take_interval": 1,
"with_depth": False,
"min_match_score": 0.80,
"min_match_score": 0.85,
"min_bbox_score": 0.25,
},
"campus": {
"path": "/datasets/campus/skelda/test.json",
"take_interval": 1,
"min_match_score": 0.89,
"min_match_score": 0.92,
"min_bbox_score": 0.5,
},
"shelf": {
"path": "/datasets/shelf/skelda/test.json",
"take_interval": 1,
"min_match_score": 0.92,
"min_match_score": 0.95,
"min_group_size": 2,
},
"ikeaasm": {
"path": "/datasets/ikeaasm/skelda/test.json",
"take_interval": 2,
"min_match_score": 0.89,
"min_match_score": 0.92,
"min_bbox_score": 0.20,
},
"chi3d": {
@ -107,21 +107,20 @@ datasets = {
"tsinghua": {
"path": "/datasets/tsinghua/skelda/test.json",
"take_interval": 3,
"min_match_score": 0.92,
"min_match_score": 0.95,
"min_group_size": 2,
},
"human36m_wb": {
"path": "/datasets/human36m/skelda/wb/test.json",
"take_interval": 100,
"min_bbox_score": 0.4,
"min_match_score": 0.93,
"batch_poses": False,
},
"egohumans_tagging": {
"path": "/datasets/egohumans/skelda/all.json",
"take_interval": 2,
"subset": "tagging",
"min_match_score": 0.89,
"min_match_score": 0.92,
"min_group_size": 2,
"min_bbox_score": 0.2,
"min_bbox_area": 0.05 * 0.05,

View File

@ -19,7 +19,7 @@ whole_body = {
"hands": False,
}
config = {
"min_match_score": 0.91,
"min_match_score": 0.94,
"min_group_size": 1,
"min_bbox_score": 0.3,
"min_bbox_area": 0.1 * 0.1,