Generative Adversarial Networks (GANs) are widely-used generative models for synthesizing complex and realistic data. However, mode collapse, where the diversity of generated samples is significantly lower than that of real samples, poses a major challenge for further applications. Our theoretical analysis demonstrates that the generator loss function is non-convex with respect to its parameters when there are multiple real modes. In particular, parameters that result in generated distributions with perfect partial mode coverage of the real distribution are the local minima of the generator loss function. To address mode collapse, we propose a unified framework called Dynamic GAN. This method detects collapsed samples in the generator by thresholding on observable discriminator outputs, divides the training set based on these collapsed samples, and trains a dynamic conditional model on the partitions. The theoretical outcome ensures progressive mode coverage and experiments on synthetic and real-world data sets demonstrate that our method surpasses several GAN variants. In conclusion, we examine the root cause of mode collapse and offer a novel approach to quantitatively detect and resolve it in GANs.

Vehicle speed estimation is an important problem in traffic surveillance. Many existing approaches to this problem are based on camera calibration. Two shortcomings exist for camera calibration-based methods. First, camera calibration methods are sensitive to the environment, which means the accuracy of the results are compromised in some situations where the environmental condition is not satisfied. Furthermore, camera calibration-based methods rely on vehicle trajectories acquired by a two-stage tracking and detection process. In an effort to overcome these shortcomings, we propose an alternate end-to-end method based on 3-dimensional convolutional networks (3D ConvNets). The proposed method bases average vehicle speed estimation on information from video footage. Our methods are characterized by the following three features. First, we use non-local blocks in our model to better capture spatial–temporal long-range dependency. Second, we use optical flow as an input in the model. Optical flow includes the information on the speed and direction of pixel motion in an image. Third, we construct a multi-scale convolutional network. This network extracts information on various characteristics of vehicles in motion. The proposed method showcases promising experimental results on commonly used dataset with mean absolute error (MAE) as $2.71$ km/h and mean square error (MSE) as $14.62$ .

Guanhua Huang, Yuchen Zhang, Zhe Li, Yongjian You, Mingze Wang, Zhouwang Yang. Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2024.