Transferring Representation to ‘Top View’ based on deep metric learning
Visualization of 'top view' by applying PCA on learned embeddings.
For inference demo:
Download test videos "4 people indoor sequence"(4p-c0.avi, 4p-c1.avi, 4p-c2.avi, 4p-c3.avi) from “EPFL” data set: Multi-camera Pedestrian Videos and detection file (4p-c0.pickle, 4p-c1.pickle, 4p-c2.pickle,4p-c3.pickle) from detection file for demo. The detection file contain detection result from detector, so that you can run without detector. Create directory ../data/train/lab, put videos and pickle files in it. run
python3 master.py
For training:
Download data and trained model from data and trained model, put them in ../box2vec/data and ../box2vec/model directory respectively