Improving object detection quality with structural constraints.

Recent researches revealed object detection networks using the simple "classification loss + localization loss" training objective are not effectively optimized in many cases, while providing additional constraints on network features could effectively improve object detection quality. Specifically, some works used constraints on training sample relations to successfully learn discriminative network features. Based on these observations, we propose Structural Constraint for improving object detection quality. Structural constraint supervises feature learning in classification and localization network branches with Fisher Loss and Equi-proportion Loss respectively, by requiring feature similarities of training sample pairs to be consistent with corresponding ground truth label similarities. Structural constraint could be applied to all object detection network architectures with the assist of our Proxy Feature design. Our experiment results showed that structural constraint mechanism is able to optimize object class instances' distribution in network feature space, and consequently detection results. Evaluations on MSCOCO2017 and KITTI datasets showed that our structural constraint mechanism is able to assist baseline networks to outperform modern counterpart detectors in terms of object detection quality.

Three-dimensional resolution and contrast-enhanced confocal microscopy with array detection.

What we believe is a novel method for improving confocal microscopy's resolution and contrast in 3D space is proposed. Based on a conventional confocal microscopy setup, we use an array detector composed of 32 photomultiplier tubes (PMTs) to replace one point-detector, where the location offset of each PMT caused a different effective point spread function (PSF). By applying array detection and the fluorescence emission difference method of an image with a solid PSF and another with a donut-shaped PSF, we can enhance lateral resolution about 27% in real time with only one scan, and improve the axial resolving ability by about 22% simultaneously. Experimental results of both fluorescent beads and living cells are presented to verify the applicability and effectiveness of our method.

Eliminating deformations in fluorescence emission difference microscopy.

We propose a method for eliminating the deformations in fluorescence emission difference microscopy (FED). Due to excessive subtraction, negative values are inevitable in the original FED method, giving rise to deformations. We propose modulating the beam to generate an extended solid focal spot and a hollow focal spot. Negative image values can be avoided by using these two types of excitation spots in FED imaging. Hence, deformations are eliminated, and the signal-to-noise ratio is improved. In deformation-free imaging, the resolution is higher than that of confocal imaging by 32%. Compared to standard FED imaging with the same level of deformations, our method provides superior resolution.

Isotropic superresolution imaging for fluorescence emission difference microscopy.

Fluorescence emission diffraction microscopy (FED) has proven to be an effective sub-diffraction-limited imaging method. In this paper, we theoretically propose a method to further enhance the resolving capability of FED. Using a coated mirror and only one objective lens, this method achieves not only the same axial resolution as 4Pi microscopy but also a higher lateral resolution. The point spread function (PSF) of our method is isotropic. According to calculations, the full width at half-maximum (FWHM) of the isotropic FED's PSF is 0.17λ along all three spatial directions. Compared with confocal microscopy, the lateral resolution is improved 0.7-fold, and the axial resolution is improved 3.1-fold. Simulation tests also demonstrate this method's advantage over traditional microscopy techniques.