2. Brief Introduction
Image big data with large volume and
high dimension.
Retrieval of images with both
computation efficiency and search
quality.
Hashing Method:
transform high-dimensional data
into compact binary codes and
generate similar binary codes for
similar data items.
3. Brief Introduction
Hashing Method
Data Independent
Data Dependent
Unsupervised Method
Supervised Method
http://cs.nju.edu.cn/lwj/L2H.html#unsupervised-hashing
4. Deep Hashing Network
•Motivation
• uncontrollable quantization error
• large approximation error by adopting ordinary distance between
continuous embeddings as the surrogate of Hamming distance between
binary codes.
Deep Hashing Network for Efficient Similarity Retrieval AAAI2016
5. Deep Hashing Network
Deep Hashing Network for Efficient Similarity Retrieval AAAI2016
(1) a sub-network with multiple convolution-pooling layers to capture image representations;
(2) a fully-connected hashing layer to generate compact binary hash codes;
(3) a pairwise cross-entropy loss layer for similarity-preserving learning;
(4) a pairwise quantization loss for controlling hashing quality.
6. Deep Hashing Network
• Bayesian Framework
• fch output [-1,1] : tanh()
• logarithm Maximum a Posteriori (MAP) estimation
•
Deep Hashing Network for Efficient Similarity Retrieval AAAI2016
7. Deep Hashing Network
• logarithm Maximum a Posteriori (MAP) estimation
•
Deep Hashing Network for Efficient Similarity Retrieval AAAI2016
13. Deep Hashing Network
• Details
• Feature layer: ReLU [0,1]
• The features extracted by the network will be very close to 0 since
they share the same scale with weights. On the contrary, our ABC
expects every dimension of the output features to be constrained near
0 or 1. As a consequence, we will encounter an unstable optimization
process if not adopting any normalization.
Adversarial Binary Coding for Efficient Person Re-identification
Unstable