Person Re-Identification using LA-Transformer.

Added some demo codes.

See other outputs of Re-id_demo.py in the Out dir. The code keep storing features in a list and when new features are detected from next frame, it calculates cosine similarity and assigns a person ID based on the maximum similarity above a certain threshold. If a new feature does not match then person id is going to be index of the new feature in the list.
The Outputs the in fliped dir are generated by not only sending a person into the model. It is generated by also sending a fliped image of the person. Adding both features will be more robust to represent the person.
In the flipped version, the model consistently tries to assign the same ID to the person wearing an orange t-shirt and cap when he first appears and later on.
Output of LA_transformer_demo.py. See how, for same person image cosine similarity is high. Persons images in imgs dir were taken from the Market-1501 dataset.
Output of person_detection.py.
The model may not work if a person changes clothes.

Requirements:

Name		Name	Last commit message	Last commit date
Latest commit History 35 Commits
Detection		Detection
LATransformer		LATransformer
Model		Model
Persons_imgs		Persons_imgs
Video		Video
.gitattributes		.gitattributes
.gitignore		.gitignore
LA-Transformer Testing.html		LA-Transformer Testing.html
LA-Transformer Testing.ipynb		LA-Transformer Testing.ipynb
LA-Transformer Training.html		LA-Transformer Training.html
LA-Transformer Training.ipynb		LA-Transformer Training.ipynb
LA_transformer_demo.py		LA_transformer_demo.py
LICENSE		LICENSE
Person_detection_demo.py		Person_detection_demo.py
README.md		README.md
Reid_demo.py		Reid_demo.py
Save_lib_v.py		Save_lib_v.py
Video_to_gif.py		Video_to_gif.py
requirements.csv		requirements.csv