Similarity Search in High Dimensions via Hashing

REQUIREMENTS

pip install -r requirements.txt

1. pytorch >= 1.0
2. loguru

DATASETS

cifar10-gist.mat password: umb6

cifar-10_alexnet.t password: f1b7

nus-wide-tc21_alexnet.t password: vfeu

imagenet-tc100_alexnet.t password: 6w5i

USAGE

usage: run.py [-h] [--dataset DATASET] [--root ROOT]
              [--code-length CODE_LENGTH] [--topk TOPK] [--gpu GPU]

LSH_PyTorch

optional arguments:
  -h, --help            show this help message and exit
  --dataset DATASET     Dataset name.
  --root ROOT           Path of dataset
  --code-length CODE_LENGTH
                        Binary hash code length.(default:
                        8,16,24,32,48,64,96,128)
  --topk TOPK           Calculate top k data map.(default: all)
  --gpu GPU             Using gpu.(default: False)

EXPERIMENTS

cifar10-gist dataset. 1000 query images, 59000 retrieval images, MAP@ALL.

cifar-10-alexnet dataset. Alexnet features, 1000 query images, 59000 retrieval images, MAP@ALL.

nus-wide-tc21-alexnet dataset. Alexnet features, top 21 classes, 2100 query images, 193734 retrieval images, MAP@5000.

imagenet-tc100-alexnet dataset. Alexnet features, top 100 classes, 5000 query images, 130000 retrieval images, MAP@1000.

Bits	8	16	24	32	48	64	96	128
cifar10-gist@ALL	0.1138	0.1191	0.1195	0.1288	0.1349	0.1436	0.1536	0.1521
cifar10-alexnet@ALL	0.1463	0.1290	0.1416	0.1588	0.1686	0.1757	0.1860	0.2121
nus-wide-tc21-alexnet@5000	0.3905	0.4632	0.4836	0.5243	0.6012	0.6051	0.6513	0.6921
imagenet-tc100-alexnet@1000	0.0536	0.0685	0.0952	0.1290	0.1861	0.2326	0.2909	0.3410