First tracker implementation.
This commit is contained in:
Daniel
591
media/RESULTS.md
591
media/RESULTS.md
@ -7,26 +7,27 @@ Results of the model in various experiments on different datasets. \
|
||||
|
||||
```json
|
||||
{
|
||||
"img_loading": 0.00946685,
|
||||
"demosaicing": 0.000433632,
|
||||
"avg_time_2d": 0.00564251,
|
||||
"avg_time_3d": 3.16907e-05,
|
||||
"fps": 163.724
|
||||
"img_loading": 0.00954196,
|
||||
"demosaicing": 0.000437263,
|
||||
"avg_time_2d": 0.00563337,
|
||||
"avg_time_3d": 3.22195e-05,
|
||||
"time_tracks": 2.56698e-06,
|
||||
"fps": 163.789
|
||||
}
|
||||
{
|
||||
"triangulator_calls": 600,
|
||||
"init_time": 2.2774e-06,
|
||||
"undistort_time": 3.66277e-06,
|
||||
"project_time": 6.4956e-07,
|
||||
"match_time": 1.27751e-06,
|
||||
"pairs_time": 1.83347e-07,
|
||||
"pair_scoring_time": 5.04129e-06,
|
||||
"grouping_time": 9.12377e-07,
|
||||
"full_time": 6.56363e-06,
|
||||
"merge_time": 2.93065e-06,
|
||||
"post_time": 3.79765e-06,
|
||||
"convert_time": 5.13667e-08,
|
||||
"total_time": 2.75941e-05
|
||||
"init_time": 2.34263e-06,
|
||||
"undistort_time": 3.69506e-06,
|
||||
"project_time": 5.953e-07,
|
||||
"match_time": 1.23644e-06,
|
||||
"pairs_time": 1.78917e-07,
|
||||
"pair_scoring_time": 5.10816e-06,
|
||||
"grouping_time": 8.91258e-07,
|
||||
"full_time": 6.67602e-06,
|
||||
"merge_time": 2.95392e-06,
|
||||
"post_time": 3.7758e-06,
|
||||
"convert_time": 5.69e-08,
|
||||
"total_time": 2.7759e-05
|
||||
}
|
||||
{
|
||||
"person_nums": {
|
||||
@ -45,23 +46,23 @@ Results of the model in various experiments on different datasets. \
|
||||
},
|
||||
"mpjpe": {
|
||||
"count": 600,
|
||||
"mean": 0.06036,
|
||||
"median": 0.052956,
|
||||
"std": 0.028032,
|
||||
"sem": 0.001145,
|
||||
"mean": 0.060543,
|
||||
"median": 0.053007,
|
||||
"std": 0.028089,
|
||||
"sem": 0.001148,
|
||||
"min": 0.037548,
|
||||
"max": 0.192385,
|
||||
"recall-0.025": 0.0,
|
||||
"recall-0.05": 0.333333,
|
||||
"recall-0.1": 0.948333,
|
||||
"recall-0.05": 0.325,
|
||||
"recall-0.1": 0.946667,
|
||||
"recall-0.15": 0.95,
|
||||
"recall-0.25": 1.0,
|
||||
"recall-0.5": 1.0,
|
||||
"num_labels": 600,
|
||||
"ap-0.025": 0.0,
|
||||
"ap-0.05": 0.187797,
|
||||
"ap-0.1": 0.910662,
|
||||
"ap-0.15": 0.913398,
|
||||
"ap-0.05": 0.179324,
|
||||
"ap-0.1": 0.907003,
|
||||
"ap-0.15": 0.91147,
|
||||
"ap-0.25": 1.0,
|
||||
"ap-0.5": 1.0
|
||||
},
|
||||
@ -99,82 +100,82 @@ Results of the model in various experiments on different datasets. \
|
||||
},
|
||||
"shoulder_right": {
|
||||
"count": 600,
|
||||
"mean": 0.045983,
|
||||
"mean": 0.046158,
|
||||
"median": 0.034992,
|
||||
"std": 0.039542,
|
||||
"sem": 0.001616,
|
||||
"std": 0.039877,
|
||||
"sem": 0.001629,
|
||||
"min": 0.00206,
|
||||
"max": 0.240597,
|
||||
"recall-0.025": 0.298333,
|
||||
"recall-0.05": 0.753333,
|
||||
"recall-0.1": 0.91,
|
||||
"recall-0.15": 0.946667,
|
||||
"recall-0.1": 0.908333,
|
||||
"recall-0.15": 0.945,
|
||||
"recall-0.25": 1.0,
|
||||
"recall-0.5": 1.0,
|
||||
"num_labels": 600
|
||||
},
|
||||
"elbow_left": {
|
||||
"count": 600,
|
||||
"mean": 0.045328,
|
||||
"median": 0.034177,
|
||||
"std": 0.03755,
|
||||
"sem": 0.001534,
|
||||
"mean": 0.045362,
|
||||
"median": 0.034249,
|
||||
"std": 0.037307,
|
||||
"sem": 0.001524,
|
||||
"min": 0.002988,
|
||||
"max": 0.194229,
|
||||
"recall-0.025": 0.26,
|
||||
"recall-0.05": 0.763333,
|
||||
"recall-0.1": 0.926667,
|
||||
"recall-0.15": 0.953333,
|
||||
"recall-0.025": 0.258333,
|
||||
"recall-0.05": 0.761667,
|
||||
"recall-0.1": 0.925,
|
||||
"recall-0.15": 0.955,
|
||||
"recall-0.25": 1.0,
|
||||
"recall-0.5": 1.0,
|
||||
"num_labels": 600
|
||||
},
|
||||
"elbow_right": {
|
||||
"count": 600,
|
||||
"mean": 0.044168,
|
||||
"mean": 0.04432,
|
||||
"median": 0.03193,
|
||||
"std": 0.03869,
|
||||
"sem": 0.001581,
|
||||
"std": 0.038646,
|
||||
"sem": 0.001579,
|
||||
"min": 0.005218,
|
||||
"max": 0.427232,
|
||||
"max": 0.383954,
|
||||
"recall-0.025": 0.268333,
|
||||
"recall-0.05": 0.783333,
|
||||
"recall-0.1": 0.928333,
|
||||
"recall-0.15": 0.941667,
|
||||
"recall-0.1": 0.926667,
|
||||
"recall-0.15": 0.94,
|
||||
"recall-0.25": 0.998333,
|
||||
"recall-0.5": 1.0,
|
||||
"num_labels": 600
|
||||
},
|
||||
"wrist_left": {
|
||||
"count": 600,
|
||||
"mean": 0.039992,
|
||||
"median": 0.022739,
|
||||
"std": 0.044805,
|
||||
"sem": 0.001831,
|
||||
"mean": 0.04031,
|
||||
"median": 0.022792,
|
||||
"std": 0.044965,
|
||||
"sem": 0.001837,
|
||||
"min": 0.002345,
|
||||
"max": 0.289452,
|
||||
"recall-0.025": 0.553333,
|
||||
"recall-0.05": 0.766667,
|
||||
"recall-0.1": 0.908333,
|
||||
"recall-0.15": 0.928333,
|
||||
"recall-0.025": 0.551667,
|
||||
"recall-0.05": 0.763333,
|
||||
"recall-0.1": 0.906667,
|
||||
"recall-0.15": 0.93,
|
||||
"recall-0.25": 0.998333,
|
||||
"recall-0.5": 1.0,
|
||||
"num_labels": 600
|
||||
},
|
||||
"wrist_right": {
|
||||
"count": 599,
|
||||
"mean": 0.040157,
|
||||
"median": 0.026117,
|
||||
"std": 0.044323,
|
||||
"sem": 0.001812,
|
||||
"count": 600,
|
||||
"mean": 0.041059,
|
||||
"median": 0.026221,
|
||||
"std": 0.04525,
|
||||
"sem": 0.001849,
|
||||
"min": 0.003181,
|
||||
"max": 0.296936,
|
||||
"recall-0.025": 0.481667,
|
||||
"recall-0.05": 0.806667,
|
||||
"recall-0.1": 0.915,
|
||||
"recall-0.15": 0.925,
|
||||
"recall-0.25": 0.995,
|
||||
"recall-0.5": 0.998333,
|
||||
"recall-0.025": 0.476667,
|
||||
"recall-0.05": 0.801667,
|
||||
"recall-0.1": 0.91,
|
||||
"recall-0.15": 0.921667,
|
||||
"recall-0.25": 0.996667,
|
||||
"recall-0.5": 1.0,
|
||||
"num_labels": 600
|
||||
},
|
||||
"hip_left": {
|
||||
@ -211,31 +212,31 @@ Results of the model in various experiments on different datasets. \
|
||||
},
|
||||
"knee_left": {
|
||||
"count": 599,
|
||||
"mean": 0.052251,
|
||||
"mean": 0.05212,
|
||||
"median": 0.042367,
|
||||
"std": 0.042631,
|
||||
"sem": 0.001743,
|
||||
"std": 0.039681,
|
||||
"sem": 0.001623,
|
||||
"min": 0.014466,
|
||||
"max": 0.482769,
|
||||
"max": 0.303977,
|
||||
"recall-0.025": 0.06,
|
||||
"recall-0.05": 0.698333,
|
||||
"recall-0.1": 0.925,
|
||||
"recall-0.15": 0.945,
|
||||
"recall-0.25": 0.995,
|
||||
"recall-0.05": 0.696667,
|
||||
"recall-0.1": 0.921667,
|
||||
"recall-0.15": 0.943333,
|
||||
"recall-0.25": 0.996667,
|
||||
"recall-0.5": 0.998333,
|
||||
"num_labels": 600
|
||||
},
|
||||
"knee_right": {
|
||||
"count": 600,
|
||||
"mean": 0.045065,
|
||||
"median": 0.034037,
|
||||
"std": 0.037143,
|
||||
"sem": 0.001518,
|
||||
"mean": 0.045269,
|
||||
"median": 0.034108,
|
||||
"std": 0.03723,
|
||||
"sem": 0.001521,
|
||||
"min": 0.010961,
|
||||
"max": 0.293895,
|
||||
"recall-0.025": 0.106667,
|
||||
"recall-0.05": 0.825,
|
||||
"recall-0.1": 0.931667,
|
||||
"recall-0.025": 0.105,
|
||||
"recall-0.05": 0.821667,
|
||||
"recall-0.1": 0.93,
|
||||
"recall-0.15": 0.948333,
|
||||
"recall-0.25": 0.998333,
|
||||
"recall-0.5": 1.0,
|
||||
@ -243,50 +244,50 @@ Results of the model in various experiments on different datasets. \
|
||||
},
|
||||
"ankle_left": {
|
||||
"count": 599,
|
||||
"mean": 0.088513,
|
||||
"median": 0.082493,
|
||||
"std": 0.029511,
|
||||
"sem": 0.001207,
|
||||
"mean": 0.089795,
|
||||
"median": 0.082637,
|
||||
"std": 0.03326,
|
||||
"sem": 0.00136,
|
||||
"min": 0.048202,
|
||||
"max": 0.271976,
|
||||
"max": 0.351142,
|
||||
"recall-0.025": 0.0,
|
||||
"recall-0.05": 0.001667,
|
||||
"recall-0.1": 0.88,
|
||||
"recall-0.15": 0.945,
|
||||
"recall-0.25": 0.995,
|
||||
"recall-0.1": 0.871667,
|
||||
"recall-0.15": 0.94,
|
||||
"recall-0.25": 0.991667,
|
||||
"recall-0.5": 0.998333,
|
||||
"num_labels": 600
|
||||
},
|
||||
"ankle_right": {
|
||||
"count": 600,
|
||||
"mean": 0.077527,
|
||||
"median": 0.06667,
|
||||
"std": 0.038806,
|
||||
"sem": 0.001586,
|
||||
"min": 0.033001,
|
||||
"mean": 0.077779,
|
||||
"median": 0.066732,
|
||||
"std": 0.039218,
|
||||
"sem": 0.001602,
|
||||
"min": 0.018484,
|
||||
"max": 0.281311,
|
||||
"recall-0.025": 0.0,
|
||||
"recall-0.05": 0.036667,
|
||||
"recall-0.1": 0.9,
|
||||
"recall-0.15": 0.926667,
|
||||
"recall-0.025": 0.003333,
|
||||
"recall-0.05": 0.041667,
|
||||
"recall-0.1": 0.893333,
|
||||
"recall-0.15": 0.923333,
|
||||
"recall-0.25": 0.991667,
|
||||
"recall-0.5": 1.0,
|
||||
"num_labels": 600
|
||||
},
|
||||
"joint_recalls": {
|
||||
"num_labels": 7800,
|
||||
"recall-0.025": 0.18962,
|
||||
"recall-0.05": 0.50128,
|
||||
"recall-0.1": 0.87962,
|
||||
"recall-0.15": 0.93526,
|
||||
"recall-0.025": 0.18897,
|
||||
"recall-0.05": 0.50064,
|
||||
"recall-0.1": 0.87718,
|
||||
"recall-0.15": 0.93423,
|
||||
"recall-0.25": 0.99603,
|
||||
"recall-0.5": 0.99923
|
||||
"recall-0.5": 0.99949
|
||||
}
|
||||
}
|
||||
{
|
||||
"total_parts": 8400,
|
||||
"correct_parts": 8134,
|
||||
"pcp": 0.968333
|
||||
"correct_parts": 8130,
|
||||
"pcp": 0.967857
|
||||
}
|
||||
```
|
||||
|
||||
@ -294,70 +295,71 @@ Results of the model in various experiments on different datasets. \
|
||||
|
||||
```json
|
||||
{
|
||||
"img_loading": 0.0431246,
|
||||
"demosaicing": 0.000535209,
|
||||
"avg_time_2d": 0.0104033,
|
||||
"avg_time_3d": 0.000107023,
|
||||
"fps": 90.5343
|
||||
"img_loading": 0.0415743,
|
||||
"demosaicing": 0.000548337,
|
||||
"avg_time_2d": 0.0104442,
|
||||
"avg_time_3d": 0.000102827,
|
||||
"time_tracks": 7.53299e-07,
|
||||
"fps": 90.1218
|
||||
}
|
||||
{
|
||||
"triangulator_calls": 301,
|
||||
"init_time": 2.60675e-06,
|
||||
"undistort_time": 1.42386e-05,
|
||||
"project_time": 1.74955e-06,
|
||||
"match_time": 7.80174e-06,
|
||||
"pairs_time": 4.016e-06,
|
||||
"pair_scoring_time": 2.31076e-05,
|
||||
"grouping_time": 4.08468e-06,
|
||||
"full_time": 2.34529e-05,
|
||||
"merge_time": 1.1163e-05,
|
||||
"post_time": 6.22243e-06,
|
||||
"convert_time": 1.01126e-07,
|
||||
"total_time": 9.87993e-05
|
||||
"init_time": 2.5757e-06,
|
||||
"undistort_time": 1.38599e-05,
|
||||
"project_time": 1.70299e-06,
|
||||
"match_time": 7.42211e-06,
|
||||
"pairs_time": 3.51909e-06,
|
||||
"pair_scoring_time": 2.20828e-05,
|
||||
"grouping_time": 4.00774e-06,
|
||||
"full_time": 2.23789e-05,
|
||||
"merge_time": 1.09032e-05,
|
||||
"post_time": 6.10427e-06,
|
||||
"convert_time": 1.11395e-07,
|
||||
"total_time": 9.49145e-05
|
||||
}
|
||||
{
|
||||
"person_nums": {
|
||||
"total_frames": 301,
|
||||
"total_labels": 477,
|
||||
"total_preds": 828,
|
||||
"total_preds": 847,
|
||||
"considered_empty": 0,
|
||||
"valid_preds": 477,
|
||||
"invalid_preds": 351,
|
||||
"invalid_preds": 370,
|
||||
"missing": 0,
|
||||
"invalid_fraction": 0.42391,
|
||||
"precision": 0.57609,
|
||||
"invalid_fraction": 0.43684,
|
||||
"precision": 0.56316,
|
||||
"recall": 1.0,
|
||||
"f1": 0.73103,
|
||||
"non_empty": 828
|
||||
"f1": 0.72054,
|
||||
"non_empty": 847
|
||||
},
|
||||
"mpjpe": {
|
||||
"count": 477,
|
||||
"mean": 0.047653,
|
||||
"mean": 0.047597,
|
||||
"median": 0.042661,
|
||||
"std": 0.015008,
|
||||
"sem": 0.000688,
|
||||
"std": 0.014947,
|
||||
"sem": 0.000685,
|
||||
"min": 0.028932,
|
||||
"max": 0.113187,
|
||||
"max": 0.107384,
|
||||
"recall-0.025": 0.0,
|
||||
"recall-0.05": 0.719078,
|
||||
"recall-0.1": 0.985325,
|
||||
"recall-0.05": 0.721174,
|
||||
"recall-0.1": 0.987421,
|
||||
"recall-0.15": 1.0,
|
||||
"recall-0.25": 1.0,
|
||||
"recall-0.5": 1.0,
|
||||
"num_labels": 477,
|
||||
"ap-0.025": 0.0,
|
||||
"ap-0.05": 0.449379,
|
||||
"ap-0.1": 0.750506,
|
||||
"ap-0.15": 0.76101,
|
||||
"ap-0.25": 0.76101,
|
||||
"ap-0.5": 0.76101
|
||||
"ap-0.05": 0.456736,
|
||||
"ap-0.1": 0.755053,
|
||||
"ap-0.15": 0.764546,
|
||||
"ap-0.25": 0.764546,
|
||||
"ap-0.5": 0.764546
|
||||
},
|
||||
"head": {
|
||||
"count": 477,
|
||||
"mean": 0.053389,
|
||||
"mean": 0.053421,
|
||||
"median": 0.049509,
|
||||
"std": 0.025237,
|
||||
"sem": 0.001157,
|
||||
"std": 0.025337,
|
||||
"sem": 0.001161,
|
||||
"min": 0.005816,
|
||||
"max": 0.182842,
|
||||
"recall-0.025": 0.098532,
|
||||
@ -402,17 +404,17 @@ Results of the model in various experiments on different datasets. \
|
||||
},
|
||||
"elbow_left": {
|
||||
"count": 477,
|
||||
"mean": 0.040654,
|
||||
"mean": 0.040136,
|
||||
"median": 0.032194,
|
||||
"std": 0.02881,
|
||||
"sem": 0.001321,
|
||||
"std": 0.026056,
|
||||
"sem": 0.001194,
|
||||
"min": 0.003888,
|
||||
"max": 0.309836,
|
||||
"max": 0.146492,
|
||||
"recall-0.025": 0.301887,
|
||||
"recall-0.05": 0.752621,
|
||||
"recall-0.1": 0.951782,
|
||||
"recall-0.15": 0.997904,
|
||||
"recall-0.25": 0.997904,
|
||||
"recall-0.1": 0.953878,
|
||||
"recall-0.15": 1.0,
|
||||
"recall-0.25": 1.0,
|
||||
"recall-0.5": 1.0,
|
||||
"num_labels": 477
|
||||
},
|
||||
@ -434,33 +436,33 @@ Results of the model in various experiments on different datasets. \
|
||||
},
|
||||
"wrist_left": {
|
||||
"count": 477,
|
||||
"mean": 0.059565,
|
||||
"mean": 0.059143,
|
||||
"median": 0.05442,
|
||||
"std": 0.039643,
|
||||
"sem": 0.001817,
|
||||
"std": 0.038602,
|
||||
"sem": 0.001769,
|
||||
"min": 0.001595,
|
||||
"max": 0.314599,
|
||||
"recall-0.025": 0.155136,
|
||||
"recall-0.05": 0.410901,
|
||||
"recall-0.1": 0.909853,
|
||||
"recall-0.15": 0.966457,
|
||||
"recall-0.25": 0.989518,
|
||||
"recall-0.1": 0.91195,
|
||||
"recall-0.15": 0.968553,
|
||||
"recall-0.25": 0.991614,
|
||||
"recall-0.5": 1.0,
|
||||
"num_labels": 477
|
||||
},
|
||||
"wrist_right": {
|
||||
"count": 477,
|
||||
"mean": 0.058378,
|
||||
"mean": 0.05783,
|
||||
"median": 0.053422,
|
||||
"std": 0.034091,
|
||||
"sem": 0.001563,
|
||||
"std": 0.030837,
|
||||
"sem": 0.001413,
|
||||
"min": 0.008434,
|
||||
"max": 0.381589,
|
||||
"max": 0.207547,
|
||||
"recall-0.025": 0.109015,
|
||||
"recall-0.05": 0.442348,
|
||||
"recall-0.1": 0.907757,
|
||||
"recall-0.15": 0.979036,
|
||||
"recall-0.25": 0.997904,
|
||||
"recall-0.15": 0.981132,
|
||||
"recall-0.25": 1.0,
|
||||
"recall-0.5": 1.0,
|
||||
"num_labels": 477
|
||||
},
|
||||
@ -514,15 +516,15 @@ Results of the model in various experiments on different datasets. \
|
||||
},
|
||||
"knee_right": {
|
||||
"count": 477,
|
||||
"mean": 0.039775,
|
||||
"mean": 0.039912,
|
||||
"median": 0.035222,
|
||||
"std": 0.023097,
|
||||
"sem": 0.001059,
|
||||
"std": 0.023393,
|
||||
"sem": 0.001072,
|
||||
"min": 0.006395,
|
||||
"max": 0.184949,
|
||||
"recall-0.025": 0.303983,
|
||||
"recall-0.05": 0.727463,
|
||||
"recall-0.1": 0.974843,
|
||||
"recall-0.1": 0.972746,
|
||||
"recall-0.15": 0.997904,
|
||||
"recall-0.25": 1.0,
|
||||
"recall-0.5": 1.0,
|
||||
@ -530,10 +532,10 @@ Results of the model in various experiments on different datasets. \
|
||||
},
|
||||
"ankle_left": {
|
||||
"count": 477,
|
||||
"mean": 0.036458,
|
||||
"mean": 0.036485,
|
||||
"median": 0.028094,
|
||||
"std": 0.03061,
|
||||
"sem": 0.001403,
|
||||
"std": 0.030728,
|
||||
"sem": 0.001408,
|
||||
"min": 0.004185,
|
||||
"max": 0.222831,
|
||||
"recall-0.025": 0.429769,
|
||||
@ -546,17 +548,17 @@ Results of the model in various experiments on different datasets. \
|
||||
},
|
||||
"ankle_right": {
|
||||
"count": 477,
|
||||
"mean": 0.040799,
|
||||
"mean": 0.041358,
|
||||
"median": 0.030877,
|
||||
"std": 0.03753,
|
||||
"sem": 0.00172,
|
||||
"std": 0.040428,
|
||||
"sem": 0.001853,
|
||||
"min": 0.002443,
|
||||
"max": 0.287278,
|
||||
"max": 0.355429,
|
||||
"recall-0.025": 0.30608,
|
||||
"recall-0.05": 0.813417,
|
||||
"recall-0.1": 0.928721,
|
||||
"recall-0.15": 0.968553,
|
||||
"recall-0.25": 0.997904,
|
||||
"recall-0.15": 0.966457,
|
||||
"recall-0.25": 0.993711,
|
||||
"recall-0.5": 1.0,
|
||||
"num_labels": 477
|
||||
},
|
||||
@ -565,15 +567,15 @@ Results of the model in various experiments on different datasets. \
|
||||
"recall-0.025": 0.21351,
|
||||
"recall-0.05": 0.62119,
|
||||
"recall-0.1": 0.94323,
|
||||
"recall-0.15": 0.98645,
|
||||
"recall-0.15": 0.98662,
|
||||
"recall-0.25": 0.99871,
|
||||
"recall-0.5": 1.0
|
||||
}
|
||||
}
|
||||
{
|
||||
"total_parts": 6678,
|
||||
"correct_parts": 6622,
|
||||
"pcp": 0.991614
|
||||
"correct_parts": 6623,
|
||||
"pcp": 0.991764
|
||||
}
|
||||
```
|
||||
|
||||
@ -581,75 +583,76 @@ Results of the model in various experiments on different datasets. \
|
||||
|
||||
```json
|
||||
{
|
||||
"img_loading": 0.00517366,
|
||||
"demosaicing": 0.000109466,
|
||||
"avg_time_2d": 0.00478839,
|
||||
"avg_time_3d": 3.38694e-05,
|
||||
"fps": 202.769
|
||||
"img_loading": 0.00477884,
|
||||
"demosaicing": 0.000105891,
|
||||
"avg_time_2d": 0.00475547,
|
||||
"avg_time_3d": 3.45411e-05,
|
||||
"time_tracks": 8.63962e-07,
|
||||
"fps": 204.216
|
||||
}
|
||||
{
|
||||
"triangulator_calls": 222,
|
||||
"init_time": 2.02099e-06,
|
||||
"undistort_time": 5.6859e-06,
|
||||
"project_time": 8.53198e-07,
|
||||
"match_time": 2.58553e-06,
|
||||
"pairs_time": 4.89811e-07,
|
||||
"pair_scoring_time": 4.32314e-06,
|
||||
"grouping_time": 8.76297e-07,
|
||||
"full_time": 5.37456e-06,
|
||||
"merge_time": 2.54044e-06,
|
||||
"post_time": 4.92114e-06,
|
||||
"convert_time": 7.91441e-08,
|
||||
"total_time": 2.99961e-05
|
||||
"init_time": 1.99938e-06,
|
||||
"undistort_time": 5.71187e-06,
|
||||
"project_time": 8.61266e-07,
|
||||
"match_time": 2.61942e-06,
|
||||
"pairs_time": 4.20405e-07,
|
||||
"pair_scoring_time": 4.35068e-06,
|
||||
"grouping_time": 9.27387e-07,
|
||||
"full_time": 5.49219e-06,
|
||||
"merge_time": 2.62484e-06,
|
||||
"post_time": 5.03611e-06,
|
||||
"convert_time": 9.22072e-08,
|
||||
"total_time": 3.03843e-05
|
||||
}
|
||||
{
|
||||
"person_nums": {
|
||||
"total_frames": 222,
|
||||
"total_labels": 376,
|
||||
"total_preds": 443,
|
||||
"total_preds": 457,
|
||||
"considered_empty": 0,
|
||||
"valid_preds": 376,
|
||||
"invalid_preds": 67,
|
||||
"invalid_preds": 81,
|
||||
"missing": 0,
|
||||
"invalid_fraction": 0.15124,
|
||||
"precision": 0.84876,
|
||||
"invalid_fraction": 0.17724,
|
||||
"precision": 0.82276,
|
||||
"recall": 1.0,
|
||||
"f1": 0.91819,
|
||||
"non_empty": 443
|
||||
"f1": 0.90276,
|
||||
"non_empty": 457
|
||||
},
|
||||
"mpjpe": {
|
||||
"count": 376,
|
||||
"mean": 0.075573,
|
||||
"median": 0.072487,
|
||||
"std": 0.017073,
|
||||
"sem": 0.000882,
|
||||
"mean": 0.075498,
|
||||
"median": 0.073144,
|
||||
"std": 0.014424,
|
||||
"sem": 0.000745,
|
||||
"min": 0.040047,
|
||||
"max": 0.237124,
|
||||
"max": 0.12107,
|
||||
"recall-0.025": 0.0,
|
||||
"recall-0.05": 0.007979,
|
||||
"recall-0.1": 0.922872,
|
||||
"recall-0.15": 0.99734,
|
||||
"recall-0.1": 0.933511,
|
||||
"recall-0.15": 1.0,
|
||||
"recall-0.25": 1.0,
|
||||
"recall-0.5": 1.0,
|
||||
"num_labels": 376,
|
||||
"ap-0.025": 0.0,
|
||||
"ap-0.05": 7.9e-05,
|
||||
"ap-0.1": 0.836487,
|
||||
"ap-0.15": 0.966449,
|
||||
"ap-0.25": 0.968931,
|
||||
"ap-0.5": 0.968931
|
||||
"ap-0.05": 9.4e-05,
|
||||
"ap-0.1": 0.843551,
|
||||
"ap-0.15": 0.959589,
|
||||
"ap-0.25": 0.959589,
|
||||
"ap-0.5": 0.959589
|
||||
},
|
||||
"head": {
|
||||
"count": 376,
|
||||
"mean": 0.063145,
|
||||
"mean": 0.063289,
|
||||
"median": 0.062777,
|
||||
"std": 0.025859,
|
||||
"sem": 0.001335,
|
||||
"std": 0.02607,
|
||||
"sem": 0.001346,
|
||||
"min": 0.007715,
|
||||
"max": 0.197424,
|
||||
"recall-0.025": 0.055851,
|
||||
"recall-0.05": 0.31117,
|
||||
"recall-0.1": 0.930851,
|
||||
"recall-0.1": 0.928191,
|
||||
"recall-0.15": 0.992021,
|
||||
"recall-0.25": 1.0,
|
||||
"recall-0.5": 1.0,
|
||||
@ -657,97 +660,97 @@ Results of the model in various experiments on different datasets. \
|
||||
},
|
||||
"shoulder_left": {
|
||||
"count": 376,
|
||||
"mean": 0.067151,
|
||||
"mean": 0.066843,
|
||||
"median": 0.064619,
|
||||
"std": 0.029971,
|
||||
"sem": 0.001548,
|
||||
"std": 0.029378,
|
||||
"sem": 0.001517,
|
||||
"min": 0.016733,
|
||||
"max": 0.205046,
|
||||
"recall-0.025": 0.039894,
|
||||
"recall-0.05": 0.321809,
|
||||
"recall-0.1": 0.890957,
|
||||
"recall-0.15": 0.981383,
|
||||
"recall-0.1": 0.893617,
|
||||
"recall-0.15": 0.984043,
|
||||
"recall-0.25": 1.0,
|
||||
"recall-0.5": 1.0,
|
||||
"num_labels": 376
|
||||
},
|
||||
"shoulder_right": {
|
||||
"count": 376,
|
||||
"mean": 0.070444,
|
||||
"mean": 0.069399,
|
||||
"median": 0.062402,
|
||||
"std": 0.038664,
|
||||
"sem": 0.001997,
|
||||
"std": 0.031883,
|
||||
"sem": 0.001646,
|
||||
"min": 0.017401,
|
||||
"max": 0.495275,
|
||||
"max": 0.198707,
|
||||
"recall-0.025": 0.018617,
|
||||
"recall-0.05": 0.316489,
|
||||
"recall-0.1": 0.837766,
|
||||
"recall-0.15": 0.968085,
|
||||
"recall-0.25": 0.99734,
|
||||
"recall-0.15": 0.970745,
|
||||
"recall-0.25": 1.0,
|
||||
"recall-0.5": 1.0,
|
||||
"num_labels": 376
|
||||
},
|
||||
"elbow_left": {
|
||||
"count": 376,
|
||||
"mean": 0.090384,
|
||||
"mean": 0.089682,
|
||||
"median": 0.072996,
|
||||
"std": 0.054936,
|
||||
"sem": 0.002837,
|
||||
"std": 0.053299,
|
||||
"sem": 0.002752,
|
||||
"min": 0.009937,
|
||||
"max": 0.344095,
|
||||
"max": 0.283516,
|
||||
"recall-0.025": 0.029255,
|
||||
"recall-0.05": 0.242021,
|
||||
"recall-0.1": 0.648936,
|
||||
"recall-0.15": 0.864362,
|
||||
"recall-0.25": 0.989362,
|
||||
"recall-0.1": 0.651596,
|
||||
"recall-0.15": 0.867021,
|
||||
"recall-0.25": 0.992021,
|
||||
"recall-0.5": 1.0,
|
||||
"num_labels": 376
|
||||
},
|
||||
"elbow_right": {
|
||||
"count": 375,
|
||||
"mean": 0.077748,
|
||||
"median": 0.070448,
|
||||
"std": 0.037755,
|
||||
"sem": 0.001952,
|
||||
"count": 376,
|
||||
"mean": 0.077757,
|
||||
"median": 0.070519,
|
||||
"std": 0.037705,
|
||||
"sem": 0.001947,
|
||||
"min": 0.004749,
|
||||
"max": 0.194529,
|
||||
"recall-0.025": 0.042553,
|
||||
"recall-0.05": 0.276596,
|
||||
"recall-0.1": 0.723404,
|
||||
"recall-0.15": 0.941489,
|
||||
"recall-0.25": 0.99734,
|
||||
"recall-0.5": 0.99734,
|
||||
"recall-0.1": 0.726064,
|
||||
"recall-0.15": 0.944149,
|
||||
"recall-0.25": 1.0,
|
||||
"recall-0.5": 1.0,
|
||||
"num_labels": 376
|
||||
},
|
||||
"wrist_left": {
|
||||
"count": 376,
|
||||
"mean": 0.128727,
|
||||
"median": 0.113244,
|
||||
"std": 0.06993,
|
||||
"sem": 0.003611,
|
||||
"mean": 0.128965,
|
||||
"median": 0.113502,
|
||||
"std": 0.068869,
|
||||
"sem": 0.003556,
|
||||
"min": 0.015136,
|
||||
"max": 0.458731,
|
||||
"max": 0.352772,
|
||||
"recall-0.025": 0.013298,
|
||||
"recall-0.05": 0.079787,
|
||||
"recall-0.1": 0.422872,
|
||||
"recall-0.15": 0.672872,
|
||||
"recall-0.25": 0.917553,
|
||||
"recall-0.1": 0.417553,
|
||||
"recall-0.15": 0.670213,
|
||||
"recall-0.25": 0.920213,
|
||||
"recall-0.5": 1.0,
|
||||
"num_labels": 376
|
||||
},
|
||||
"wrist_right": {
|
||||
"count": 376,
|
||||
"mean": 0.104259,
|
||||
"median": 0.09734,
|
||||
"std": 0.050125,
|
||||
"sem": 0.002588,
|
||||
"mean": 0.102037,
|
||||
"median": 0.097009,
|
||||
"std": 0.044155,
|
||||
"sem": 0.00228,
|
||||
"min": 0.002425,
|
||||
"max": 0.461713,
|
||||
"max": 0.277303,
|
||||
"recall-0.025": 0.005319,
|
||||
"recall-0.05": 0.093085,
|
||||
"recall-0.1": 0.515957,
|
||||
"recall-0.15": 0.853723,
|
||||
"recall-0.25": 0.984043,
|
||||
"recall-0.05": 0.095745,
|
||||
"recall-0.1": 0.521277,
|
||||
"recall-0.15": 0.864362,
|
||||
"recall-0.25": 0.994681,
|
||||
"recall-0.5": 1.0,
|
||||
"num_labels": 376
|
||||
},
|
||||
@ -769,17 +772,17 @@ Results of the model in various experiments on different datasets. \
|
||||
},
|
||||
"hip_right": {
|
||||
"count": 376,
|
||||
"mean": 0.065474,
|
||||
"mean": 0.065421,
|
||||
"median": 0.059695,
|
||||
"std": 0.035357,
|
||||
"sem": 0.001826,
|
||||
"std": 0.034174,
|
||||
"sem": 0.001765,
|
||||
"min": 0.005283,
|
||||
"max": 0.302623,
|
||||
"max": 0.259007,
|
||||
"recall-0.025": 0.053191,
|
||||
"recall-0.05": 0.359043,
|
||||
"recall-0.1": 0.906915,
|
||||
"recall-0.15": 0.976064,
|
||||
"recall-0.25": 0.989362,
|
||||
"recall-0.1": 0.904255,
|
||||
"recall-0.15": 0.973404,
|
||||
"recall-0.25": 0.992021,
|
||||
"recall-0.5": 1.0,
|
||||
"num_labels": 376
|
||||
},
|
||||
@ -801,15 +804,15 @@ Results of the model in various experiments on different datasets. \
|
||||
},
|
||||
"knee_right": {
|
||||
"count": 376,
|
||||
"mean": 0.053573,
|
||||
"median": 0.049537,
|
||||
"std": 0.02731,
|
||||
"sem": 0.00141,
|
||||
"mean": 0.053806,
|
||||
"median": 0.049906,
|
||||
"std": 0.027608,
|
||||
"sem": 0.001426,
|
||||
"min": 0.008841,
|
||||
"max": 0.230822,
|
||||
"recall-0.025": 0.087766,
|
||||
"recall-0.05": 0.50266,
|
||||
"recall-0.1": 0.954787,
|
||||
"recall-0.05": 0.5,
|
||||
"recall-0.1": 0.952128,
|
||||
"recall-0.15": 0.992021,
|
||||
"recall-0.25": 1.0,
|
||||
"recall-0.5": 1.0,
|
||||
@ -817,32 +820,32 @@ Results of the model in various experiments on different datasets. \
|
||||
},
|
||||
"ankle_left": {
|
||||
"count": 376,
|
||||
"mean": 0.072656,
|
||||
"median": 0.066373,
|
||||
"std": 0.046322,
|
||||
"sem": 0.002392,
|
||||
"mean": 0.074289,
|
||||
"median": 0.066969,
|
||||
"std": 0.045644,
|
||||
"sem": 0.002357,
|
||||
"min": 0.009841,
|
||||
"max": 0.390551,
|
||||
"max": 0.378337,
|
||||
"recall-0.025": 0.042553,
|
||||
"recall-0.05": 0.329787,
|
||||
"recall-0.1": 0.861702,
|
||||
"recall-0.15": 0.962766,
|
||||
"recall-0.25": 0.984043,
|
||||
"recall-0.05": 0.327128,
|
||||
"recall-0.1": 0.832447,
|
||||
"recall-0.15": 0.944149,
|
||||
"recall-0.25": 0.989362,
|
||||
"recall-0.5": 1.0,
|
||||
"num_labels": 376
|
||||
},
|
||||
"ankle_right": {
|
||||
"count": 376,
|
||||
"mean": 0.063864,
|
||||
"median": 0.052121,
|
||||
"std": 0.048194,
|
||||
"sem": 0.002489,
|
||||
"mean": 0.066324,
|
||||
"median": 0.052238,
|
||||
"std": 0.048596,
|
||||
"sem": 0.002509,
|
||||
"min": 0.007344,
|
||||
"max": 0.373408,
|
||||
"recall-0.025": 0.069149,
|
||||
"recall-0.05": 0.43883,
|
||||
"recall-0.1": 0.896277,
|
||||
"recall-0.15": 0.960106,
|
||||
"recall-0.05": 0.430851,
|
||||
"recall-0.1": 0.859043,
|
||||
"recall-0.15": 0.946809,
|
||||
"recall-0.25": 0.986702,
|
||||
"recall-0.5": 1.0,
|
||||
"num_labels": 376
|
||||
@ -850,17 +853,17 @@ Results of the model in various experiments on different datasets. \
|
||||
"joint_recalls": {
|
||||
"num_labels": 4888,
|
||||
"recall-0.025": 0.04194,
|
||||
"recall-0.05": 0.30176,
|
||||
"recall-0.1": 0.80135,
|
||||
"recall-0.15": 0.93269,
|
||||
"recall-0.25": 0.98732,
|
||||
"recall-0.5": 0.99959
|
||||
"recall-0.05": 0.30115,
|
||||
"recall-0.1": 0.79705,
|
||||
"recall-0.15": 0.93208,
|
||||
"recall-0.25": 0.98977,
|
||||
"recall-0.5": 1.0
|
||||
}
|
||||
}
|
||||
{
|
||||
"total_parts": 5264,
|
||||
"correct_parts": 4995,
|
||||
"pcp": 0.948898
|
||||
"correct_parts": 5005,
|
||||
"pcp": 0.950798
|
||||
}
|
||||
```
|
||||
|
||||
|
||||
289
rpt/tracker.hpp
Normal file
289
rpt/tracker.hpp
Normal file
@ -0,0 +1,289 @@
|
||||
#pragma once
|
||||
|
||||
#include <array>
|
||||
#include <string>
|
||||
#include <vector>
|
||||
#include <algorithm>
|
||||
#include <limits>
|
||||
#include <cmath>
|
||||
#include <iostream>
|
||||
|
||||
// =================================================================================================
|
||||
|
||||
struct Track
|
||||
{
|
||||
std::vector<std::vector<std::array<float, 4>>> core_poses;
|
||||
std::vector<std::vector<std::array<float, 4>>> full_poses;
|
||||
|
||||
std::vector<double> timestamps;
|
||||
size_t id;
|
||||
};
|
||||
|
||||
// =================================================================================================
|
||||
|
||||
class PoseTracker
|
||||
{
|
||||
public:
|
||||
PoseTracker(float max_distance);
|
||||
|
||||
std::vector<std::tuple<size_t, std::vector<std::array<float, 4>>>> track_poses(
|
||||
const std::vector<std::vector<std::array<float, 4>>> &poses_3d,
|
||||
const std::vector<std::string> &joint_names,
|
||||
const double timestamp);
|
||||
|
||||
void reset();
|
||||
|
||||
private:
|
||||
float max_distance;
|
||||
size_t history_size = 3;
|
||||
float max_movement_speed = 2.0;
|
||||
|
||||
std::vector<double> timestamps;
|
||||
std::vector<Track> pose_tracks;
|
||||
|
||||
const std::vector<std::string> core_joints = {
|
||||
"shoulder_left",
|
||||
"shoulder_right",
|
||||
"hip_left",
|
||||
"hip_right",
|
||||
"elbow_left",
|
||||
"elbow_right",
|
||||
"knee_left",
|
||||
"knee_right",
|
||||
"wrist_left",
|
||||
"wrist_right",
|
||||
"ankle_left",
|
||||
"ankle_right",
|
||||
};
|
||||
|
||||
int match_to_track(const std::vector<std::array<float, 4>> &core_pose_3d);
|
||||
|
||||
std::vector<std::array<float, 4>> refine_pose(const Track &track);
|
||||
};
|
||||
|
||||
// =================================================================================================
|
||||
// =================================================================================================
|
||||
|
||||
PoseTracker::PoseTracker(float max_distance)
|
||||
{
|
||||
this->max_distance = max_distance;
|
||||
}
|
||||
|
||||
// =================================================================================================
|
||||
|
||||
void PoseTracker::reset()
|
||||
{
|
||||
pose_tracks.clear();
|
||||
timestamps.clear();
|
||||
}
|
||||
|
||||
// =================================================================================================
|
||||
|
||||
std::vector<std::tuple<size_t, std::vector<std::array<float, 4>>>> PoseTracker::track_poses(
|
||||
const std::vector<std::vector<std::array<float, 4>>> &poses_3d,
|
||||
const std::vector<std::string> &joint_names,
|
||||
const double timestamp)
|
||||
{
|
||||
// Extract core joints
|
||||
std::vector<size_t> core_joint_idx;
|
||||
for (const auto &joint : core_joints)
|
||||
{
|
||||
auto it = std::find(joint_names.begin(), joint_names.end(), joint);
|
||||
core_joint_idx.push_back(std::distance(joint_names.begin(), it));
|
||||
}
|
||||
std::vector<std::vector<std::array<float, 4>>> core_poses;
|
||||
core_poses.resize(poses_3d.size());
|
||||
for (size_t i = 0; i < poses_3d.size(); ++i)
|
||||
{
|
||||
core_poses[i].resize(core_joint_idx.size());
|
||||
for (size_t j = 0; j < core_joint_idx.size(); ++j)
|
||||
{
|
||||
for (size_t k = 0; k < 4; ++k)
|
||||
{
|
||||
core_poses[i][j][k] = poses_3d[i][core_joint_idx[j]][k];
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
// Match core poses to tracks
|
||||
for (size_t i = 0; i < core_poses.size(); ++i)
|
||||
{
|
||||
int track_idx = match_to_track(core_poses[i]);
|
||||
if (track_idx == -1)
|
||||
{
|
||||
// Create a new track
|
||||
Track new_track;
|
||||
new_track.core_poses.push_back(core_poses[i]);
|
||||
new_track.full_poses.push_back(poses_3d[i]);
|
||||
new_track.timestamps.push_back(timestamp);
|
||||
new_track.id = pose_tracks.size();
|
||||
pose_tracks.push_back(new_track);
|
||||
}
|
||||
else
|
||||
{
|
||||
// Update existing track
|
||||
auto &track = pose_tracks[track_idx];
|
||||
track.core_poses.push_back(core_poses[i]);
|
||||
track.full_poses.push_back(poses_3d[i]);
|
||||
track.timestamps.push_back(timestamp);
|
||||
}
|
||||
}
|
||||
|
||||
// Remove old tracks
|
||||
timestamps.push_back(timestamp);
|
||||
if (timestamps.size() > history_size)
|
||||
{
|
||||
timestamps.erase(timestamps.begin());
|
||||
}
|
||||
double max_age = timestamps.front();
|
||||
for (size_t i = 0; i < pose_tracks.size();)
|
||||
{
|
||||
auto &track = pose_tracks[i];
|
||||
double last_timestamp = track.timestamps.back();
|
||||
if (last_timestamp < max_age)
|
||||
{
|
||||
pose_tracks.erase(pose_tracks.begin() + i);
|
||||
}
|
||||
else
|
||||
{
|
||||
++i;
|
||||
}
|
||||
}
|
||||
|
||||
// Remove old poses from tracks
|
||||
for (auto &track : pose_tracks)
|
||||
{
|
||||
while (track.core_poses.size() > history_size)
|
||||
{
|
||||
track.core_poses.erase(track.core_poses.begin());
|
||||
track.full_poses.erase(track.full_poses.begin());
|
||||
track.timestamps.erase(track.timestamps.begin());
|
||||
}
|
||||
}
|
||||
|
||||
// Refine poses
|
||||
std::vector<std::tuple<size_t, std::vector<std::array<float, 4>>>> tracked_poses;
|
||||
for (size_t i = 0; i < pose_tracks.size(); ++i)
|
||||
{
|
||||
auto &track = pose_tracks[i];
|
||||
if (track.core_poses.size() > 0)
|
||||
{
|
||||
std::vector<std::array<float, 4>> refined_pose = refine_pose(track);
|
||||
tracked_poses.emplace_back(track.id, refined_pose);
|
||||
}
|
||||
}
|
||||
|
||||
return tracked_poses;
|
||||
}
|
||||
|
||||
// =================================================================================================
|
||||
|
||||
int PoseTracker::match_to_track(const std::vector<std::array<float, 4>> &core_pose_3d)
|
||||
{
|
||||
int best_track = -1;
|
||||
float best_distance_sq = max_distance * max_distance;
|
||||
|
||||
for (size_t i = 0; i < pose_tracks.size(); ++i)
|
||||
{
|
||||
const auto &track = pose_tracks[i];
|
||||
if (track.core_poses.size() == 0)
|
||||
continue;
|
||||
|
||||
// Calculate distance to the last pose in the track
|
||||
const auto &last_pose = track.core_poses.back();
|
||||
float distance_sq = 0.0;
|
||||
for (size_t j = 0; j < core_pose_3d.size(); ++j)
|
||||
{
|
||||
float dx = core_pose_3d[j][0] - last_pose[j][0];
|
||||
float dy = core_pose_3d[j][1] - last_pose[j][1];
|
||||
float dz = core_pose_3d[j][2] - last_pose[j][2];
|
||||
distance_sq += dx * dx + dy * dy + dz * dz;
|
||||
}
|
||||
distance_sq /= core_pose_3d.size();
|
||||
|
||||
if (distance_sq < best_distance_sq)
|
||||
{
|
||||
best_distance_sq = distance_sq;
|
||||
best_track = static_cast<int>(i);
|
||||
}
|
||||
}
|
||||
return best_track;
|
||||
}
|
||||
|
||||
// =================================================================================================
|
||||
|
||||
std::vector<std::array<float, 4>> PoseTracker::refine_pose(const Track &track)
|
||||
{
|
||||
// Restrict maximum movement by physical constraints, by clipping the pose to the maximum
|
||||
// movement distance from one of the track's history poses
|
||||
//
|
||||
// While advanced sensor filtering techniques, like using a Kalman-Filter, might improve the
|
||||
// average accuracy of the pose, they introduce update delays on fast movement changes. For
|
||||
// example, if a person stands still for a while and then suddenly moves, the first few frames
|
||||
// will likely be treated as outliers, since the filter will not be able to adapt fast enough.
|
||||
// This behaviour is not desired in a real-time critical applications, where the pose needs to
|
||||
// be updated to the real physical position of the person as fast as possible. Therefore, only
|
||||
// the movement speed is limited here.
|
||||
|
||||
if (track.core_poses.size() < 2)
|
||||
{
|
||||
return track.full_poses.back();
|
||||
}
|
||||
|
||||
// Calculate maximum possible movement distance from each history pose
|
||||
std::vector<float> max_movement_dists_sq;
|
||||
max_movement_dists_sq.resize(track.core_poses.size() - 1);
|
||||
double last_timestamp = track.timestamps.back();
|
||||
for (size_t i = 0; i < track.core_poses.size() - 1; ++i)
|
||||
{
|
||||
double ts = track.timestamps[i];
|
||||
float max_movement = max_movement_speed * (last_timestamp - ts);
|
||||
max_movement_dists_sq[i] = max_movement * max_movement;
|
||||
}
|
||||
|
||||
// Clip joint if required
|
||||
std::vector<std::array<float, 4>> refined_pose = track.full_poses.back();
|
||||
for (size_t i = 0; i < refined_pose.size(); ++i)
|
||||
{
|
||||
float min_dist_sq = std::numeric_limits<float>::infinity();
|
||||
size_t closest_idx = 0;
|
||||
bool clip = true;
|
||||
|
||||
for (size_t j = 0; j < max_movement_dists_sq.size(); ++j)
|
||||
{
|
||||
|
||||
float dx = refined_pose[i][0] - track.full_poses[j][i][0];
|
||||
float dy = refined_pose[i][1] - track.full_poses[j][i][1];
|
||||
float dz = refined_pose[i][2] - track.full_poses[j][i][2];
|
||||
float dist_sq = dx * dx + dy * dy + dz * dz;
|
||||
if (dist_sq < min_dist_sq)
|
||||
{
|
||||
min_dist_sq = dist_sq;
|
||||
closest_idx = j;
|
||||
}
|
||||
if (dist_sq <= max_movement_dists_sq[j])
|
||||
{
|
||||
clip = false;
|
||||
break;
|
||||
}
|
||||
}
|
||||
|
||||
if (clip)
|
||||
{
|
||||
float dist_sq = min_dist_sq;
|
||||
float scale = max_movement_dists_sq[closest_idx] / dist_sq;
|
||||
|
||||
float dx = refined_pose[i][0] - track.full_poses[closest_idx][i][0];
|
||||
float dy = refined_pose[i][1] - track.full_poses[closest_idx][i][1];
|
||||
float dz = refined_pose[i][2] - track.full_poses[closest_idx][i][2];
|
||||
refined_pose[i][0] = track.full_poses[closest_idx][i][0] + dx * scale;
|
||||
refined_pose[i][1] = track.full_poses[closest_idx][i][1] + dy * scale;
|
||||
refined_pose[i][2] = track.full_poses[closest_idx][i][2] + dz * scale;
|
||||
|
||||
// Set confidence to a low value if the joint is clipped
|
||||
refined_pose[i][3] = 0.1;
|
||||
}
|
||||
}
|
||||
|
||||
return refined_pose;
|
||||
}
|
||||
@ -17,6 +17,7 @@ using json = nlohmann::json;
|
||||
|
||||
#include "/RapidPoseTriangulation/rpt/camera.hpp"
|
||||
#include "/RapidPoseTriangulation/rpt/interface.hpp"
|
||||
#include "/RapidPoseTriangulation/rpt/tracker.hpp"
|
||||
#include "/RapidPoseTriangulation/scripts/utils_2d_pose.hpp"
|
||||
#include "/RapidPoseTriangulation/scripts/utils_pipeline.hpp"
|
||||
|
||||
@ -121,6 +122,7 @@ int main(int argc, char **argv)
|
||||
const float min_match_score = config["min_match_score"];
|
||||
const size_t min_group_size = config["min_group_size"];
|
||||
const int take_interval = config["take_interval"];
|
||||
const float ifps = config["fps"];
|
||||
|
||||
// Load 2D model
|
||||
bool use_wb = utils_pipeline::use_whole_body(whole_body);
|
||||
@ -131,6 +133,9 @@ int main(int argc, char **argv)
|
||||
// Load 3D model
|
||||
std::unique_ptr<Triangulator> tri_model = std::make_unique<Triangulator>(
|
||||
min_match_score, min_group_size);
|
||||
const float max_distance = 0.3 + 2.0 / ifps;
|
||||
std::unique_ptr<PoseTracker> pose_tracker = std::make_unique<PoseTracker>(
|
||||
max_distance);
|
||||
|
||||
// Timers
|
||||
size_t time_count = dataset.size();
|
||||
@ -138,10 +143,12 @@ int main(int argc, char **argv)
|
||||
std::vector<double> times_debayer;
|
||||
std::vector<double> times_pose2d;
|
||||
std::vector<double> times_pose3d;
|
||||
std::vector<double> times_tracks;
|
||||
times_image.reserve(time_count);
|
||||
times_debayer.reserve(time_count);
|
||||
times_pose2d.reserve(time_count);
|
||||
times_pose3d.reserve(time_count);
|
||||
times_tracks.reserve(time_count);
|
||||
size_t print_steps = (size_t)std::floor((float)time_count / 100.0f);
|
||||
print_steps = std::max((size_t)1, print_steps);
|
||||
|
||||
@ -214,6 +221,7 @@ int main(int argc, char **argv)
|
||||
{
|
||||
// Reset last poses if scene changes
|
||||
tri_model->reset();
|
||||
pose_tracker->reset();
|
||||
old_scene = item["scene"];
|
||||
}
|
||||
|
||||
@ -241,6 +249,19 @@ int main(int argc, char **argv)
|
||||
elapsed = std::chrono::high_resolution_clock::now() - stime;
|
||||
times_pose3d.push_back(elapsed.count());
|
||||
|
||||
stime = std::chrono::high_resolution_clock::now();
|
||||
double ts = ((int)item["index"]) / ifps;
|
||||
auto pose_tracks = pose_tracker->track_poses(poses_3d, joint_names_2d, ts);
|
||||
std::vector<std::vector<std::array<float, 4>>> poses_3d_refined;
|
||||
for (size_t j = 0; j < pose_tracks.size(); j++)
|
||||
{
|
||||
auto &pose = std::get<1>(pose_tracks[j]);
|
||||
poses_3d_refined.push_back(pose);
|
||||
}
|
||||
poses_3d = poses_3d_refined;
|
||||
elapsed = std::chrono::high_resolution_clock::now() - stime;
|
||||
times_tracks.push_back(elapsed.count());
|
||||
|
||||
all_poses_3d.push_back(std::move(poses_3d));
|
||||
all_ids.push_back(item["id"].get<std::string>());
|
||||
old_id = item["index"];
|
||||
@ -254,23 +275,27 @@ int main(int argc, char **argv)
|
||||
double time_debayer = 0.0;
|
||||
double time_pose2d = 0.0;
|
||||
double time_pose3d = 0.0;
|
||||
double time_tracks = 0.0;
|
||||
for (size_t i = warmup; i < time_count; i++)
|
||||
{
|
||||
time_image += times_image[i];
|
||||
time_debayer += times_debayer[i];
|
||||
time_pose2d += times_pose2d[i];
|
||||
time_pose3d += times_pose3d[i];
|
||||
time_tracks += times_tracks[i];
|
||||
}
|
||||
double avg_time_image = time_image / (time_count - warmup);
|
||||
double avg_time_debayer = time_debayer / (time_count - warmup);
|
||||
double avg_time_pose2d = time_pose2d / (time_count - warmup);
|
||||
double avg_time_pose3d = time_pose3d / (time_count - warmup);
|
||||
double fps = 1.0 / (avg_time_debayer + avg_time_pose2d + avg_time_pose3d);
|
||||
double avg_time_tracks = time_tracks / (time_count - warmup);
|
||||
double fps = 1.0 / (avg_time_debayer + avg_time_pose2d + avg_time_pose3d + avg_time_tracks);
|
||||
std::cout << "{\n"
|
||||
<< " \"img_loading\": " << avg_time_image << ",\n"
|
||||
<< " \"demosaicing\": " << avg_time_debayer << ",\n"
|
||||
<< " \"avg_time_2d\": " << avg_time_pose2d << ",\n"
|
||||
<< " \"avg_time_3d\": " << avg_time_pose3d << ",\n"
|
||||
<< " \"time_tracks\": " << avg_time_tracks << ",\n"
|
||||
<< " \"fps\": " << fps << "\n"
|
||||
<< "}" << std::endl;
|
||||
tri_model->print_stats();
|
||||
|
||||
@ -55,6 +55,7 @@ datasets = {
|
||||
"human36m": {
|
||||
"path": "/datasets/human36m/skelda/pose_test.json",
|
||||
"take_interval": 5,
|
||||
"fps": 50,
|
||||
"min_match_score": 0.95,
|
||||
"min_group_size": 1,
|
||||
"min_bbox_score": 0.4,
|
||||
@ -84,6 +85,7 @@ datasets = {
|
||||
},
|
||||
"campus": {
|
||||
"path": "/datasets/campus/skelda/test.json",
|
||||
"fps": 20,
|
||||
"take_interval": 1,
|
||||
"min_match_score": 0.92,
|
||||
"min_bbox_score": 0.5,
|
||||
@ -91,6 +93,7 @@ datasets = {
|
||||
"shelf": {
|
||||
"path": "/datasets/shelf/skelda/test.json",
|
||||
"take_interval": 1,
|
||||
"fps": 20,
|
||||
"min_match_score": 0.95,
|
||||
"min_group_size": 2,
|
||||
},
|
||||
@ -346,6 +349,7 @@ def main():
|
||||
"batch_poses": batch_poses,
|
||||
"whole_body": whole_body,
|
||||
"take_interval": datasets[dataset_use]["take_interval"],
|
||||
"fps": datasets[dataset_use]["fps"],
|
||||
}
|
||||
utils_pipeline.save_json(config, config_path)
|
||||
|
||||
|
||||
Reference in New Issue
Block a user