Personnel tracking: https://github.com/layumi/Person_reID_baseline_pytorch
What 0.pf_extractor.extract_feature inside 751?
Market-1501 people inside dataset 751
What 1.tracker inside self.cur_list preservation is?
Stored in the current frame box each person: p.pf features, p.pbox block, p.pscore: the inner product with a score that best matches the feature box, p.track_status: 1,
I on a box with no corresponding box, the new box, is None
I.e. it is stored and an associated upper information box
2.pre_thresh = 0.85, person_thresh = 0.7 all mean?
pre_thresh means that the current frame box i, and an upper box j, the inner product of the result of the determination threshold value for the same individual
person_thresh, the coincidence of the current frame which is not an expert, who has been tracked with similar threshold
2.5 matr = pfs_new.mm (pfs_list.t ()) What do these words?
pfs_new inside the current frame, coincides with others who less than 40%.
pfs_list all people have tracked, remove the current high degree of coincidence of people that do not coincide with others who have tracked
There are a few frames person is not detected, the re-emergence
3.self.get_max input and return value all mean?
Input is a similarity matrix of nxm, the current frame and the previous frame n m,
Output res [i] = j, the current i-th frame people, and on a matching of people
pscore [i], is the i, j matching score
4.matr = pfs.mm (pfs_pre_list.t ()) This matrix multiplication What does it mean?
pfs, n individual features identifying the current picture nx751, m pfs_pre_list on an individual picture features identified mx715, after multiplication of nxm, ij i and j element represents similarity al
5._loop () What is the overall flow inside the reaction
a) detect reid pedestrian detection feature b) give each of the small picture pedestrian c) characterized by reid, person box, make video tracking d) for each person box face detecting human e) updating personlist, face information f) using human face correction g) del_and_fusion treated repeater
Reference using Kalman filter tracking mode:
https://towardsdatascience.com/object-detection-and-tracking-in-pytorch-b3cf1a696a98
