README.md

BagTrick with PCA reduction

reduce feature dimension with PCA, obtaining shorter feature meanwhile keep strong accuracy.

run

# train
python train.py --config_file ./base_config.yaml

# infer
python infer.py --config_file ./base_config.yaml --model_path /path/to/model.pth

Experimental Results and Trained Models

Settings (on a MacBook Pro (Retina, 13-inch, Mid 2014))

• GPU: TITAN XP (memory 12194MB)
• CPU: 2.6 GHz Dual-Core Intel Core i5
• Memory: 8 GB 1600 MHz DDR3

DukeMTMC-reID

DukeMTMC -ReID (gallery size: 17661)	light-reid				performance					time(on a TITAN XP)
	light model	light feature	light search	reduction	CNN	feature dim	metric	R1	mAP	inference per batch(64)	search per query
1	-	-	-	none	ResNet50	2048	cosine	0.870	0.772	78.6ms	237.1ms
2	-	-	-	pca-128	ResNet50	128	cosine	0.863	0.752	78.6ms	20.7ms

Market-1501

• comming soon

MSMT17

• comming soon