Controlling object motion trajectories in Text-to-Video (T2V) generation is a challenging and relatively under-explored area, particularly in scenarios involving multiple moving objects. Most community models and datasets in the T2V domain are designed for single-object motion, limiting the performance of current generative models in multi-object tasks. Additionally, existing motion control methods in T2V either lack support for multi-object motion scenes or experience severe performance degradation when object trajectories intersect, primarily due to the semantic conflicts in colliding regions. To address these limitations, we introduce LayerT2V, the first approach for generating video by compositing background and foreground objects layer by layer. This layered generation enables flexible integration of multiple independent elements within a video, positioning each element on a distinct “layer” and thus facilitating coherent multi-object synthesis while enhancing control over the generation process. Extensive experiments demonstrate the superiority of LayerT2V in generating complex multi-object scenarios, showcasing 1.4× and 4.5× improvements in mIoU and AP50 metrics over state-of-the-art (SOTA) methods. The code will be made publicly available.

We introduce a novel bootstrapping approach for training generative models. Specifically, we construct synthetic datasets by combining generated samples from previous iterations with real data. By recycling samples over successive generations, this technique reduces the dependence on large curated datasets while producing varied outputs.

Using information theoretic metrics for node pruning to learn the importance of individual neurons at different levels in the whole DNN. Entropy, Mutual information and KL-Selectivity are used to determine the order of ablation.

This is the course project I designed on my own for CS2107 where I served as TA, which is basically generated from Graphics And Mixed Environemnt Seminar, Lingqi Yan, UCSB, but chooses to use a more modern programming language, i.e. Rust. The project includes 3 LABs, simply allows students to learn the basics of rasterization in graphics and, most importantly, to have fun.

We employ the Machine Unlearning approach to mitigate the retention of unethical data within LLMs and prevent the generation of harmful responses. We carefully design a method to ensure: (1) For a negative Q&A training pair, the LLM forgets its original response to the input; (2) The LLM randomly maps negative prompts to any output distribution within its output space; (3) The LLM maintains a level of general language ability close to its original state post-unlearning.