🎮
👋 Hi, there! I’m Yiming Ma (é©¬ä¸€é“ in Chinese). As a PhD candidate at MathSys CDT at the University of Warwick, my research focuses on multimodality in computer vision and its applications (e.g., in crowd counting and driver monitoring systems). I am passionate about bridging mathematics and deep learning to solve real-world problems.
Skills
80%
50%
95%
Featured Publications
State-of-the-art crowd counting models follow an encoder-decoder approach. Images are first processed by the encoder to extract features. Then, to account for perspective distortion, the highest-level feature map is fed to extra components to extract multiscale features, which are the input to the decoder to generate crowd densities. However, in these methods, features extracted at earlier stages during encoding are underutilised, and the multiscale modules can only capture a limited range of receptive fields, albeit with considerable computational cost. This paper proposes a novel crowd counting architecture (FusionCount), which exploits the adaptive fusion of a large majority of encoded features instead of relying on additional extraction components to obtain multiscale features. Thus, it can cover a more extensive scope of receptive field sizes and lower the computational cost. We also introduce a new channel reduction block, which can extract saliency information during decoding and further enhance the model’s performance. Experiments on two benchmark databases demonstrate that our model achieves state-of-the-art results with reduced computational complexity. PyTorch implementation of the model and weights trained on these two datasets are available at https://github.com/YimingMa/FusionCount.
Recent Publications
Experience
This research projects aims to build an intelligent interior sensing systems for monitoring and recognising drivers’ activities from heterogeneous multi-view, multimodal data, encompassing video, audio, heart rate and driving trends. Efficient deep learning models will be developed to combine information from disparate modalities, detect relevant activities and accurately classify the detected activity. Additionally, the project will answer critical research questions including
- Which class of deep models is the best suited for the activity recognition task?
- Which subset or combination of these modalities will provide the most efficient means to monitor driver activities?
- What are the limitations of the SOTA models?
Responsibilities include:
- Literature reviewing;
- Data accessing;
- Model designing.
Education
- Supervisors: Dr. Victor Sanchez & Dr. Tanaya Guha.
- Memo: The first half year was an extension of my MSc individual project. We proposed an efficient multi-scale method that leverages extracted features directly and does not require extra modules to handle head size variation. Then funded by Ford, we moved towards multi-view multimodal driver monitoring systems, which exploit various cameras (e.g., RGB, IR, and depth) deployed at different in-car positions (e.g., on the top of the driver) to monitor the driver’s activities. We proposed a multi-branch model that employs the multi-head self-attention mechanism to fuse extracted feature maps. Our current focus is the general multimodal interaction/integration methods with applications including but not restricted to crowd counting and driver monitoring systems.
- Modules: Data Analysis and Machine Learning, Stochastic Modelling and Random Processes, Numerical Algorithms and Optimisation, Fundamentals of Mathematical Modelling, Data Mining, Agent-Based Systems, Interdisciplinary Approaches to Machine Learning.
- Group Project: Prediction of Oestrus Intervals for Guide Dogs, supervised by Prof. Colm Connaughton. This project is about predicting the next oestrus intervals of guide dogs based on features such as breeds, ages, foods, etc.
- Individual Project: Digital Twins of Urban Crowds, supervised by: Dr. Victor Sanchez & Dr. Tanaya Guha. This project aims to utilise a single traffic camera feed to develop an end-to-end deep network that estimates the crowd density in a given scene in near real-time.
- Modules (GPA: 92.18/100):
- Pure Maths: Mathematical Analysis, Linear Algebra, Complex Analysis, Real Analysis, ODE, PDE, Functional Analysis, Fourier Analysis, etc.
- Applied Maths: Numerical Analysis, Optimisation, etc.
- Statistics: Mathematical Statistics, Linear Statistical Models, Time Series Analysis, etc.
- Thesis: Contraction Methods in Composite Optimisation, supervised by Prof. Bingsheng He.
- Awards: the 1st (2019), 2nd (2017), 3rd (2018) Class of the Merit Student Scholarships.