Boundary-aware Backward-Compatible Representation via Adversarial Learning in Image Retrieval
Authors:
Tan Pan,
Furong Xu,
Xudong Yang,
Sifeng He,
Chen Jiang,
Qingpei Guo,
Feng Qian Xiaobo Zhang,
Yuan Cheng,
Lei Yang,
Wei Chu
Abstract:
Image retrieval plays an important role in the Internet world. Usually, the core parts of mainstream visual retrieval systems include an online service of the embedding model and a large-scale vector database. For traditional model upgrades, the old model will not be replaced by the new one until the embeddings of all the images in the database are re-computed by the new model, which takes days or…
▽ More
Image retrieval plays an important role in the Internet world. Usually, the core parts of mainstream visual retrieval systems include an online service of the embedding model and a large-scale vector database. For traditional model upgrades, the old model will not be replaced by the new one until the embeddings of all the images in the database are re-computed by the new model, which takes days or weeks for a large amount of data. Recently, backward-compatible training (BCT) enables the new model to be immediately deployed online by making the new embeddings directly comparable to the old ones. For BCT, improving the compatibility of two models with less negative impact on retrieval performance is the key challenge. In this paper, we introduce AdvBCT, an Adversarial Backward-Compatible Training method with an elastic boundary constraint that takes both compatibility and discrimination into consideration. We first employ adversarial learning to minimize the distribution disparity between embeddings of the new model and the old model. Meanwhile, we add an elastic boundary constraint during training to improve compatibility and discrimination efficiently. Extensive experiments on GLDv2, Revisited Oxford (ROxford), and Revisited Paris (RParis) demonstrate that our method outperforms other BCT methods on both compatibility and discrimination. The implementation of AdvBCT will be publicly available at https://github.com/Ashespt/AdvBCT.
△ Less
Submitted 4 May, 2023;
originally announced May 2023.
Proton radiography of magnetic field produced by ultra-intense laser irradiating capacity-coil target
Authors:
W. W. Wang,
J. Teng,
J. Chen,
H. B. Cai,
S. K. He,
W. M. Zhou,
L. Q. Shan,
F. Lu,
Y. C. Wu,
W. Hong,
D. X. Liu,
B. Bi,
F. Zhang,
F. B. Xue,
B. Y. Li,
B. Zhang,
Y. L. He,
W. He,
J. L. Jiao,
K. G. Dong,
F. Q. Zhang,
Z. G. Deng,
Z. M. Zhang,
B. Cui,
D. Han
, et al. (7 additional authors not shown)
Abstract:
Ultra-intense ultra-short laser is firstly used to irradiate the capacity-coil target to generate magnetic field. The spatial structure and temporal evolution of huge magnetic fields were studied with time-gated proton radiography method. A magnetic flux density of 40T was measured by comparing the proton deflection and particle track simulations. Although the laser pulse duration is only 30fs, th…
▽ More
Ultra-intense ultra-short laser is firstly used to irradiate the capacity-coil target to generate magnetic field. The spatial structure and temporal evolution of huge magnetic fields were studied with time-gated proton radiography method. A magnetic flux density of 40T was measured by comparing the proton deflection and particle track simulations. Although the laser pulse duration is only 30fs, the generated magnetic field can last for over 100 picoseconds. The energy conversion efficiency from laser to magnetic field can reach as high as ~20%. The results indicate that tens of tesla (T) magnetic field could be produced in many ultra intense laser facilities around the world, and higher magnetic field could be produced by picosecond lasers.
△ Less
Submitted 17 November, 2014;
originally announced November 2014.
