Feature Learning based Deep Supervised Hashing with Pairwise Labels

REQUIREMENTS

1. pytorch
2. loguru

pip install -r requirements.txt

DATASETS

1. CIFAR-10
2. NUS-WIDE Password: uhr3
3. Imagenet100 Password: ynwf

USAGE

usage: run.py [-h] [--dataset DATASET] [--root ROOT] [--num-query NUM_QUERY]
              [--arch ARCH] [--num-train NUM_TRAIN]
              [--code-length CODE_LENGTH] [--topk TOPK] [--gpu GPU] [--lr LR]
              [--batch-size BATCH_SIZE] [--max-iter MAX_ITER]
              [--num-workers NUM_WORKERS]
              [--evaluate-interval EVALUATE_INTERVAL] [--eta ETA]

DPSH_PyTorch

optional arguments:
  -h, --help            show this help message and exit
  --dataset DATASET     Dataset name.
  --root ROOT           Path of dataset
  --num-query NUM_QUERY
                        Number of query data points.(default: 1000)
  --arch ARCH           CNN model name.(default: alexnet)
  --num-train NUM_TRAIN
                        Number of training data points.(default: 5000)
  --code-length CODE_LENGTH
                        Binary hash code length.(default: 12,24,32,48)
  --topk TOPK           Calculate map of top k.(default: all)
  --gpu GPU             Using gpu.(default: False)
  --lr LR               learning rate(default: 1e-5)
  --batch-size BATCH_SIZE
                        batch size(default: 128)
  --max-iter MAX_ITER   Number of iterations.(default: 150)
  --num-workers NUM_WORKERS
                        Number of loading data threads.(default: 6)
  --evaluate-interval EVALUATE_INTERVAL
                        Evaluation interval(default: 10)
  --eta ETA             Hyper-parameter.(default: 0.1)

EXPERIMENTS

CNN model: Alexnet. Compute mean average precision(MAP).

cifar10: 1000 query images, 5000 training images.

nus-wide-tc21: 21 classes, 2100 query images, 10500 training images.

imagenet100: 100 classes, 5000 query images, 10000 training images.

bits	12	16	24	32	48	64	128
cifar10@ALL	0.6676	0.7131	0.7118	0.7362	0.7487	0.7542	0.7565
nus-wide-tc21@5000	0.8091	0.8188	0.8346	0.8403	0.8450	0.8503	0.8588
imagenet100@1000	0.1985	0.2497	0.3654	0.4147	0.4612	0.4950	0.5687