Qiongxiong Ma

Dehao Huang, Jintao Cheng, Rui Fan et al.

For a long time, bone marrow cell morphology examination has been an essential tool for diagnosing blood diseases. However, it is still mainly dependent on the subjective diagnosis of experienced doctors, and there is no objective quantitative standard. Therefore, it is crucial to study a robust bone marrow cell detection algorithm for a quantitative automatic analysis system. Currently, due to the dense distribution of cells in the bone marrow smear and the diverse cell classes, the detection of bone marrow cells is difficult. The existing bone marrow cell detection algorithms are still insufficient for the automatic analysis system of bone marrow smears. This paper proposes a bone marrow cell detection algorithm based on the YOLOv5 network, trained by minimizing a novel loss function. The classification method of bone marrow cell detection tasks is the basis of the proposed novel loss function. Since bone marrow cells are classified according to series and stages, part of the classes in adjacent stages are similar. The proposed novel loss function considers the similarity between bone marrow cell classes, increases the penalty for prediction errors between dissimilar classes, and reduces the penalty for prediction errors between similar classes. The results show that the proposed loss function effectively improves the algorithm's performance, and the proposed bone marrow cell detection algorithm has achieved better performance than other cell detection algorithms.

Yu Runsheng, Shi Zhenyu, Ma Qiongxiong et al.

The unsupervised Pretraining method has been widely used in aiding human action recognition. However, existing methods focus on reconstructing the already present frames rather than generating frames which happen in future.In this paper, We propose an improved Variantial Autoencoder model to extract the features with a high connection to the coming scenarios, also known as Predictive Learning. Our framework lists as following: two steam 3D-convolution neural networks are used to extract both spatial and temporal information as latent variables. Then a resample method is introduced to create new normal distribution probabilistic latent variables and finally, the deconvolution neural network will use these latent variables generate next frames. Through this possess, we train the model to focus more on how to generate the future and thus it will extract the future high connected features. In the experiment stage, A large number of experiments on UT and UCF101 datasets reveal that future generation aids Prediction does improve the performance. Moreover, the Future Representation Learning Network reach a higher score than other methods when in half observation. This means that Future Representation Learning is better than the traditional Representation Learning and other state- of-the-art methods in solving the human action prediction problems to some extends.