Deep lens optimization has recently emerged as a new paradigm for designing computational imaging systems, however it has been limited to either simple optical systems consisting of a single DOE or metalens, or the fine-tuning of compound lenses from good initial designs. Here we present a deep lens design method based on curriculum learning, which is able to learn optical designs of compound lenses ab initio from randomly initialized surfaces, therefore overcoming the need for a good initial design. We demonstrate this approach with the fully-automatic design of an extended depth-of-field computational camera in a cellphone-style form factor, highly aspherical surfaces, and a short back focal length.

Curriculum Learning for ab initio Deep Learned Refractive Optics

This paper tackles spectral reflectance recovery (SRR) from RGB images. Since capturing ground-truth spectral reflectance and camera spectral sensitivity are challenging and costly, most existing approaches are trained on synthetic images and utilize the same parameters for all unseen testing images, which are suboptimal especially when the trained models are tested on real images because they never exploit the internal information of the testing images. To address this issue, we adopt a self-supervised meta-auxiliary learning (MAXL) strategy that fine-tunes the well-trained network parameters with each testing image to combine external with internal information. To the best of our knowledge, this is the first work that successfully adapts the MAXL strategy to this problem. Instead of relying on naive end-to-end training, we also propose a novel architecture that integrates the physical relationship between the spectral reflectance and the corresponding RGB images into the network based on our mathematical analysis. Besides, since the spectral reflectance of a scene is independent to its illumination while the corresponding RGB images are not, we recover the spectral reflectance of a scene from its RGB images captured under multiple illuminations to further reduce the unknown. Qualitative and quantitative evaluations demonstrate the effectiveness of our proposed network and of the MAXL. Our code and data are available at https://github.com/Dong-Huo/SRR-MAXL.

Learning to Recover Spectral Reflectance from RGB Images

In computer vision, it has long been taken for granted that high-quality images obtained through well-designed camera lenses would lead to superior results. However, we find that this common perception is not a"one-size-fits-all"solution for diverse computer vision tasks. We demonstrate that task-driven and deep-learned simple optics can actually deliver better visual task performance. The Task-Driven lens design approach, which relies solely on a well-trained network model for supervision, is proven to be capable of designing lenses from scratch. Experimental results demonstrate the designed image classification lens (``TaskLens'') exhibits higher accuracy compared to conventional imaging-driven lenses, even with fewer lens elements. Furthermore, we show that our TaskLens is compatible with various network models while maintaining enhanced classification accuracy. We propose that TaskLens holds significant potential, particularly when physical dimensions and cost are severely constrained.

Image Quality Is Not All You Want: Task-Driven Lens Design for Image Classification

Diffractive snapshot hyperspectral imaging based on the deep optics framework has been striving to capture the spectral images of dynamic scenes. However, existing deep optics frameworks all suffer from the mismatch between the optical hardware and the reconstruction algorithm due to the quantization operation in the diffractive optical element (DOE) fabrication, leading to the limited performance of hyperspectral imaging in practice. In this paper, we propose the quantization-aware deep optics for diffractive snapshot hyperspectral imaging. Our key observation is that common lithography techniques used in fabricating DOEs need to quantize the DOE height map to a few levels, and can freely set the height for each level. Therefore, we propose to integrate the quantization operation into the DOE height map optimization and design an adaptive mechanism to adjust the physical height of each quantization level. According to the optimization, we fabricate the quantized DOE directly and build a diffractive hyperspectral snapshot imaging system. Our method develops the deep optics framework to be more practical through the awareness of and adaptation to the quantization operation of the DOE physical structure, making the fabricated DOE and the reconstruction algorithm match each other systematically. Extensive synthetic simulation and real hardware experiments validate the superior performance of our method.

Quantization-aware Deep Optics for Diffractive Snapshot Hyperspectral Imaging

Background: The extent to which maternal antibodies against the hepatitis B surface antigen (HBsAb) acquired transplacentally affect the immune responses to the hepatitis B vaccine (HBVac) in infants is still uncertain. Objective: To explore the impact of the HBsAb on the immune response to the HBVac in a mouse model. Methods: According to the doses of the HBVac (2, 5 μg) injected, 267 BALB/c mice were divided into two groups. Each group was subdivided into 3 subgroups based on the doses of the hepatitis B immunoglobulin (HBIG) (0, 25, 50 IU) administered. The HBsAb titers were detected 4 weeks after completing the HepB vaccination. Results: Among all the mice, 40 had an HBsAb titer <100 mIU/mL (non- or low-response to the HBVac). The rates of the HBsAb titer <100 mIU/mL in 0, 25 and 50 IU HBIG groups were 1.1%, 23.1%, and 20.7%, respectively. Multivariate logistic regression analysis showed that the risk factors for low- or non-response to the HBVac were injection with the HBIG, low HBVac dose, and hypodermic injection. The mean HBsAb titers (log10) reduced gradually in the 0, 25 and 50 IU HBIG groups (P<0.001). Conclusion: The HBIG administration has negative impacts on the peak level of the HBsAb and the rate of an effective immune response. This implies that the maternal HBsAb acquired transplacentally might inhibit the immune responses to the HBVac in infants.

Exploring the Influence of Hepatitis B Immunoglobulin on the Immune Response to the Hepatitis B Vaccine in a Mouse Model.

In this paper, facing the high-contrast, high-resolution and high-sensitivity technical difficulties that need to be solvedfordirect detection of habitable exoplanets, we research on the high-precision simulation method of spatial distributedsynthetic aperture optical interference system, which is one of the main ways to achieve this detection goal, by ZEMAXsoftware. We have built the simulation model of optical system for Michelson-type synthetic aperture telescope. It isverified from the principle and simulation analysis in ZEMAX software that the simulation precision of the optical system can achieve a high-contrast better than 10-7, and the simulation results of the distribution and variationof theinterference fringes are correct. Finally, we analyze the influence of factors such as band width and optical systemerroron the key index of the fringe visibility. The research in this paper can provide the necessary foundationfor theparameter decomposition and optimization design of the future spatial distributed synthetic aperture optical interferencesystem in habitable exoplanets.

Research on the high-precision simulation method of spatial distributed synthetic aperture optical interference system

There was an accidental death of BALB/c mice in the course of the experiment about hepatitis B immunization. The reasons of the mice breeding failure were analyzed in order to provide some experience for future animal preparation about hepatitis B immunization. Items of all mice including sex, age, dose of hepatitis B vaccine (HepBVac), dose of hepatitis B immunoglobulin (HBIG), immunization route, vaccination schedule and the acclimatization period were recorded before the first immunization. Among 334 mice included, the survival rates in different groups of initial ages were 45.9% in 3-week-old, 49.3% in 4-week-old, 51.9% in 5-week-old, 84.8% in 7-week-old and 96.6% in 8-week-old; the survival rate in seven days acclimatization group was 96.6% and 53.3% in one day acclimatization group. Multivariate logistic regression analysis indicated that smaller ages (3 weeks, 4 weeks) and acclimatization for only one day were the independent risk factors affecting the survival of mice. Multivariate logistic regression analysis showed that injection with HBIG, lower dose of HepBVac and hypodermic injection were independent risk factors for low-and non-response to HepB Vac. It is suggested that the mice should be more than five-week-old and must have acclimated to the environment for one week before the experiment begins. The initial ages of the mice have no impact on their immune efficacy.

	Key words: BALB/c mice, age, acclimatization, hepatitis B immunization.

/pdf/factors-influencing-survival-of-balb-c-mice-and-their-7fvq2248km.pdf

Ling Li

Papers

Quantization-aware Deep Optics for Diffractive Snapshot Hyperspectral Imaging

Exploring the Influence of Hepatitis B Immunoglobulin on the Immune Response to the Hepatitis B Vaccine in a Mouse Model.

Research on the high-precision simulation method of spatial distributed synthetic aperture optical interference system

Factors influencing survival of BALB/c mice and their effects on immune efficacy in an experiment of hepatitis B immunization