Search

The Image Signal Processor (ISP) is a customized device to restore RGB images from the pixel signals of CMOS image sensor. In order to realize this function, a series of processing units are leveraged to tackle different artifacts, such as color shifts, signal noise, moire effects, and so on, that are introduced from the photo-capturing devices. However, tuning each processing unit is highly complicated and requires a lot of experience and effort from image experts. In this paper, a novel network architecture, CSANet, with emphases on inference speed and high PSNR is proposed for end-to-end learned ISP task. The proposed CSANet applies a double attention module employing both channel and spatial attentions. Particularly, its spatial attention is simplified to a light-weighted dilated depth-wise convolution and still performs as well as others. As proof of performance, CSANet won 2nd place in the Mobile AI 2021 Learned Smartphone ISP Challenge with 1st place PSNR score.

As the quality of mobile cameras starts to play a crucial role in modern smartphones, more and more attention is now being paid to ISP algorithms used to improve various perceptual aspects of mobile photos. In this Mobile AI challenge, the target was to develop an end-to-end deep learning-based image signal processing (ISP) pipeline that can replace classical hand-crafted ISPs and achieve nearly real-time performance on smartphone NPUs. For this, the participants were provided with a novel learned ISP dataset consisting of RAW-RGB image pairs captured with the Sony IMX586 Quad Bayer mobile sensor and a professional 102-megapixel medium format camera. The runtime of all models was evaluated on the MediaTek Dimensity 1000+ platform with a dedicated AI processing unit capable of accelerating both floating-point and quantized neural networks. The proposed solutions are fully compatible with the above NPU and are capable of processing Full HD photos under 60-100 milliseconds while achieving high fidelity results. A detailed description of all models developed in this challenge is provided in this paper., Mobile AI 2021 Workshop and Challenges: https://ai-benchmark.com/workshops/mai/2021/

Co-activations of agonist and antagonist muscles are believed to be present in voluntary limb movement. Recent studies indicate that such co-activations are either synergic or dyssynergic. The aims of this paper are to (1) develop a novel method that can extract both the intensity and frequency information from the recordings of the surface electromyograms (EMGs) of involved muscles, and (2) investigate if the involved muscles will be under synergic co-activation during voluntary forearm pronation for normal subjects and dyssynergic co-activation for patients with radial nerve palsy. We examined 11 healthy subjects and 4 patients with right-arm radial nerve palsy in this study. For the group of healthy subjects, each one of them was asked to perform 30 trials of voluntary forearm pronation and then 30 trials of passive pronation as control experiments. As to the second group of patients, each one was asked to perform only 15 trials of voluntary pronation due to the limitation and durability of their arms. The recordings of the surface EMGs included the short and long heads of the biceps brachii, the brachialis, the lateral head of the triceps brachii, brachioradialis, and pronator teres. Experimental results of the healthy group indicated that the surface EMGs of all muscles had no statistically significant changes in fractal dimensions (FDs) and spectral frequencies of the control experiments during passive pronation. Yet, during the voluntary pronation experiments, the surface EMGs of all muscle groups were temporally synchronized in frequencies with persistent intensities. Hence, all involved muscle groups were in synergic co-activation. Statistical results of the group mean values of FDs during rest vs. forearm pronation also revealed significant difference with p < 0.01 for healthy subjects. As to the group of patients, their EMGs could still have bursting activities, but the synchronized significant frequencies might be lacking or the intensities as indicated from their FDs would not be persistent. To further compare the FDs among the three different protocols, a mixed-model ANOVA and multiple comparison tests were performed. Finally, in order to illustrate the advantages of this novel method, we have compared it with the detrended fluctuation analysis (DFA). It is believed that this proposed method will have the potential to be a biomarker for evaluating dynamical disease in neuromuscular disorders.

In reflex and volitional actions, co-activations of agonist and antagonist muscles are believed to be present. Recent studies indicate that such co-activations can be either synergic or dyssynergic. The aim of this paper is to investigate if the co-activations of biceps brachii, brachialis, and triceps brachii during volitional elbow flexion are in the synergic or dyssynergic state. In this study, two groups with each containing six healthy male volunteers participated. Each person of the first group performed 30 trials of volitional elbow flexion while each of the second group performed 30 trials of passive elbow flexion as control experiments. Based on the model of fractional Brownian motion, the intensity and frequency information of the surface electromyograms (EMGs) could be extracted simultaneously. No statistically significant changes were found in the control group. As to the other group, results indicated that the surface EMGs of all five muscle groups were temporally synchronized in frequencies with persistent intensities during each elbow flexion. In addition, the mean values of fractal dimensions for rest and volitional flexion states revealed significant differences with P0.01. The obtained positive results suggest that these muscle groups work together synergically to facilitate elbow flexion during the co-activations.