A5089449284- Advanced Image and Video Retrieval Techniques
- Video Surveillance and Tracking Methods
- Statistical and numerical algorithms
- Advanced Vision and Imaging
- Time Series Analysis and Forecasting
- Advanced Neural Network Applications
- Robotics and Sensor-Based Localization
- Autonomous Vehicle Technology and Safety
- Visual perception and processing mechanisms
- Advanced Optical Sensing Technologies
- Remote-Sensing Image Classification
- Photonic and Optical Devices
- Optical Coherence Tomography Applications
- Forecasting Techniques and Applications
Valeo (France)
2022-2024
Transformers have proven superior performance for a wide variety of tasks since they were introduced. In recent years, drawn attention from the vision community in such as image classification and object detection. Despite this wave, an accurate efficient multiple-object tracking (MOT) method based on transformers is yet to be designed. We argue that direct application transformer architecture with quadratic complexity insufficient noise-initialized sparse queries – not optimal MOT. propose...
Light fields capture 3D scene information by recording light rays emitted from a at various orientations. They offer more immersive perception, compared with classic 2D images, but the cost of huge data volumes. In this paper, we design compact neural network representation for field compression task. same vein as deep image prior, takes randomly initialized noise input and is trained in supervised manner order to best reconstruct target Sub-Aperture Images (SAIs). The composed two types...
In autonomous driving, motion prediction aims at forecasting the future trajectories of nearby agents, helping ego vehicle to anticipate behaviors and drive safely. A key challenge is generating a diverse set predictions, commonly addressed using data-driven models with Multiple Choice Learning (MCL) architectures Winner-Takes-All (WTA) training objectives. However, these methods face initialization sensitivity instabilities. Additionally, compensate for limited performance, some approaches...
Motion forecasting (MF) for autonomous driving aims at anticipating trajectories of surrounding agents in complex urban scenarios. In this work, we investigate a mixed strategy MF training that first pre-train motion forecasters on pseudo-labeled data, then fine-tune them annotated data. To obtain trajectories, propose simple pipeline leverages off-the-shelf single-frame 3D object detectors and non-learning trackers. The whole pre-training including pseudo-labeling is coined as PPT. Our...
Light field is a type of image data that captures the 3D scene information by recording light rays emitted from at various orientations. It offers more immersive perception than classic 2D images but cost huge volume. In this paper, we draw inspiration visual characteristics Sub-Aperture Images (SAIs) and design compact neural network representation for compression task. The backbone takes randomly initialized noise as input supervised on SAIs target field. composed two types complementary...