visual_mot_distance_dist.py

import numpy as np
import matplotlib.pyplot as plt

from tsbench.algo import cal_dist

tracks_file = "dataset/DanceTrack/sample/0004.txt"
gt_tracks = np.loadtxt(tracks_file, delimiter=",").astype(int)


tracks_file = "output/MOTDataset/DanceTrack/TDTR_IOU/0.2_iou/0004.txt"
ts_tracks = np.loadtxt(tracks_file, delimiter=",").astype(int)

tids = np.unique(ts_tracks[:, 1])

ref_ratio = 0.01
for tid in tids:
    gt_track = gt_tracks[gt_tracks[:, 1] == tid]
    ts_track = ts_tracks[ts_tracks[:, 1] == tid]

    wh = gt_track[:, 4:6]
    gt_track[:, 4:6] += gt_track[:, 2:4]
    ts_track[:, 4:6] += ts_track[:, 2:4]

    iou_dist = 1 - cal_dist.ious(gt_track[:, 2:6], ts_track[:, 2:6])
    iou_dist = iou_dist * 100

    lt_dist = np.linalg.norm(gt_track[:, 2:4] - ts_track[:, 2:4], axis=-1)
    rb_dist = np.linalg.norm(gt_track[:, 4:6] - ts_track[:, 4:6], axis=-1)
    pt_dist = (lt_dist + rb_dist) / 2

    ref_size = ref_ratio * np.linalg.norm(wh, axis=1)
    lt_ref_dist = lt_dist / ref_size
    rb_ref_dist = rb_dist / ref_size
    pt_ref_dist = (lt_ref_dist + rb_ref_dist) / 2

    # x = np.arange(gt_track.shape[0])
    x = np.arange(100)
    plt.plot(x, iou_dist[:100], label="iou_dist")
    plt.plot(x, pt_dist[:100], label="pt_dist")
    plt.plot(x, pt_ref_dist[:100], label="pt_ref_dist")
    plt.legend()
    plt.show()
    assert False