DeepHash: Getting Regularization, Depth and Fine-Tuning Right

By Jie Lin, Olivier Morère, Vijay Ramaseshan Chandrasekhar, Antoine Veillard, Hanlin Goh. 2015

35 citations recorded by the source corpus.

This work focuses on representing very high-dimensional global image descriptors using very compact 64-1024 bit binary hashes for instance retrieval. We propose DeepHash: a hashing scheme based on deep networks. Key to making DeepHash work at extremely low bitrates are three important considerations -- regularization, depth and fine-tuning -- each requiring solutions specific to the hashing problem. In-depth evaluation shows that our scheme consistently outperforms state-of-the-art methods across all data sets for both Fisher Vectors and Deep Convolutional Neural Network features, by up to 20 percent over other schemes. The retrieval performance with 256-bit hashes is close to that of the uncompressed floating point features -- a remarkable 512 times compression.

Subfields: Computer Vision and Pattern Recognition

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Vision and Pattern Recognition

1. Luc Van Gool
2. Ming-Hsuan Yang
3. Dacheng Tao
4. G. Wang
5. Yi Yang
6. Trevor Darrell
7. Jianfei Cai
8. Jitendra Malik

Citations in the corpus

• Learning Spread-Out Local Feature Descriptors (2017, background): “As shown in [9], this often leads to overfitting.”
• Recent advances in convolutional neural networks (2015, background): “[48] find that such a single margin loss function causes a dramatic drop in retrieval results when fine-tuning the network on all pairs.”
• Localizing and Orienting Street Views Using Overhead Imagery (2016, background): “One way to alleviate this problem is to add another margin to the contrastive loss function to cut the loss when the distance is small enough [18].”
• Fully Convolutional Geometric Features (2019, background): “Lin et al. [19] showed that the constraints for positive pairs could lead to overfitting and proposed a margin-based loss for positive pairs: where I ij = 1 if ( i, j ) ∈ P and 0 otherwise, and ¯ · is the NOT operator. m p and m n are margins for positive and negative pairs.”
• Learning Deep Embeddings with Histogram Loss (2016, background): “…the fact that attempting to collapse all positive pairs may lead to excessive overfitting and therefore suggested losses that mitigate this effect, e.g. a double-margin contrastive loss [12], which drops to zero for positive pairs as long as their distances fall beyond the second (smaller) margin.”