Predictable Dual-view Hashing
MOHAMMAD RASTEGARI JONGHYUN CHOI SHOBEIR FAKHRAEI* HAL DAUMÉ III LARRY S. DAVIS
University of Maryland, College Park, MD, USA
Dual-view Hashing MOHAMMAD RASTEGARI JONGHYUN CHOI SHOBEIR - - PowerPoint PPT Presentation
Predictable Dual-view Hashing MOHAMMAD RASTEGARI JONGHYUN CHOI SHOBEIR FAKHRAEI* HAL DAUM III LARRY S. DAVIS University of Maryland, College Park, MD, USA Integrate information from different modalities Text - Image Human Riding
MOHAMMAD RASTEGARI JONGHYUN CHOI SHOBEIR FAKHRAEI* HAL DAUMÉ III LARRY S. DAVIS
University of Maryland, College Park, MD, USA
ICML 2013 | Atlanta | Predictable Dual-view Hashing
Integrate information from different modalities
Human Riding Horse
ICML 2013 | Atlanta | Predictable Dual-view Hashing
Dual-View Hashing
Textual Space Visual Space
Human riding a horse Person laughing at restaurant Cow standing in a farm
010111010 101010101 110010101 Hamming Space
ICML 2013 | Atlanta | Predictable Dual-view Hashing
1 1 1 001 101
Binary code assignments with splits
ICML 2013 | Atlanta | Predictable Dual-view Hashing
5 10 15 20 25
2 4 6 8 10 12 14
01?101
Predictability
ICML 2013 | Atlanta | Predictable Dual-view Hashing
5 10 15 20 25
2 4 6 8 10 12 14
100110 010110 110110 010011 01?101
Predictability
ICML 2013 | Atlanta | Predictable Dual-view Hashing
Predictability
5 10 15 20 25
2 4 6 8 10 12 14
CCA
ICML 2013 | Atlanta | Predictable Dual-view Hashing
Predictability
5 10 15 20 25
2 4 6 8 10 12 14
PDH CCA
margin
ICML 2013 | Atlanta | Predictable Dual-view Hashing
Optimization
Solved by a block coordinate descent algorithm
ICML 2013 | Atlanta | Predictable Dual-view Hashing
Predictable dual-view hashing
1101 0010 1010 0101 1110 0101 1110 1010 0101 1001
label Modality 2 Modality 1
ICML 2013 | Atlanta | Predictable Dual-view Hashing
Predictable dual-view hashing
1101 0010 1010 0101 1110 0101 1110 1010 0101 1001
label Modality 2 Modality 1
ICML 2013 | Atlanta | Predictable Dual-view Hashing
Image image (with 32 bits)
ICML 2013 | Atlanta | Predictable Dual-view Hashing
Text Image (with 32 bits)
ICML 2013 | Atlanta | Predictable Dual-view Hashing
Image Retrieval
Pascal Sentence [Farhadi et al] Sun Attribute [Peterson et al]
Mohammad Rastegari, Jonghyun Choi, Shobeir Fakhraei*, Hal Daumé III and Larry S. Davis
University of Maryland, College Park, MD, USA