annaduring / dhn_pytorch

Source code for paper "Deep Hashing Network for Efficient Similarity Retrieval" on AAAI-2016

Deep Hashing Network for Efficient Similarity Retrieval

REQUIREMENTS

pip install -r requirements.txt

1. pytorch >= 1.0
2. loguru

DATASETS

1. CIFAR-10 Password: aemd
2. NUS-WIDE Password: msfv
3. Imagenet100 Password: xpab

USAGE

usage: run.py [-h] [--dataset DATASET] [--root ROOT]
              [--code-length CODE_LENGTH] [--arch ARCH]
              [--batch-size BATCH_SIZE] [--lr LR] [--max-iter MAX_ITER]
              [--num-workers NUM_WORKERS] [--topk TOPK] [--gpu GPU]
              [--lamda LAMDA] [--seed SEED]
              [--evaluate-interval EVALUATE_INTERVAL]

DHN_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.
  --arch ARCH           CNN model name.(default: alexnet)
  --batch-size BATCH_SIZE
                        Batch size.(default: 256)
  --lr LR               Learning rate.(default: 1e-5)
  --max-iter MAX_ITER   Number of iterations.(default: 500)
  --num-workers NUM_WORKERS
                        Number of loading data threads.(default: 6)
  --topk TOPK           Calculate map of top k.(default: all)
  --gpu GPU             Using gpu.(default: False)
  --lamda LAMDA         Hyper-parameter.(default: 1)
  --seed SEED           Random seed.(default: 3367)
  --evaluate-interval EVALUATE_INTERVAL
                        Evaluation interval.(default: 10)

EXPERIMENTS

CNN model: Alexnet.

cifar10: 1000 query images, 5000 training images, MAP@ALL.

nus-wide: Top 21 classes, 2100 query images, 10500 training images, MAP@5000.

imagenet100: Top 100 classes, 5000 query images, 10000 training images, MAP@1000.

bits	16	32	48	128
cifar10@ALL	0.7275	0.7353	0.7302	0.7386
nus-wide-tc21@5000	0.8194	0.8326	0.8396	0.8443
imagenet100@1000	0.2659	0.3703	0.4122	0.4743