Research Article
Improved Deep Hashing with Scalable Interblock for Tourist Image Retrieval
Algorithm 1
Key steps of the approach.
| Input: , the training image matrix | | q, the hash code length | | j, number of sub-layers | | W, the weight matrix | | Output: deep hash functions h(x) | (1) | Initialize the deep models by the pre-trained VGG-16 Sub-Network | (2) | Update W in training process according to loss function; | (3) | For do | (4) | For iter = 1 to j do | (5) | Compute ; | (6) | Compute ; | (7) | Quantize the relaxation binary into binary code with ; | (8) | Return | (9) | End for | (10) | End for |
|