nhquanqt / hcmus-person-reid

Final project of Computer Vision

Person Re-identification

Final project of Computer Vision

Table of content

• Person Re-identification
  • Table of content
  • Students:
  • Proposed method
  • Dataset
  • Dataset preparation
    • Download raw dataset
    • Set dataset path
  • Train
    • Config
    • Run
  • Evaluation

Students:

Name	Student ID	Email
Nguyen Hoang Quan	18120145	[e-mail]
Le Minh Khoa	18120415	[e-mail]

Proposed method

Dataset

• Market-1501

Dataset preparation

Download raw dataset

Download Market-1501 via Google Drive: link

Set dataset path

In config.json file, change root value to dataset path.

Train

Config

Check config.json file for training configuration.

Run

Run python main.py to start training.

Evaluation

Methods	Train feature	Test feature	Rank-1	Rank-5	mAP
Baseline	Global	Global	83.28	94.18	65.96
Baseline	Local	Local	80.20	92.76	59.66
Baseline	Global, Local	Global	84.14	93.97	67.58
Baseline	Global, Local	Local	86.37	95.10	68.68
Baseline + Identity Loss	Global, Local	Global	83.55	93.35	67.66
Baseline + Identity Loss	Global, Local	Local	84.23	93.62	63.99
Baseline + Identity Loss + Random Flip	Global, Local	Global	85.42	93.97	69.34
Baseline + Identity Loss + Random Flip	Global, Local	Local	86.82	94.89	67.59
Re-ranking	Global, Local	Global	87.47	92.96	82.90
Re-ranking + Identity Loss	Global, Local	Global	86.19	92.76	81.27
Re-ranking + Identity loss + Random Flip	Global, Local	Global	87.50	92.93	82.68