OPV2V: An Open Benchmark Dataset and Fusion Pipeline for Perception with Vehicle-to-Vehicle Communication

V2XP-ASG: Generating Adversarial Scenes for Vehicle-to-Everything Perception

“…V2X Perception: V2X perception investigates how to leverage the visual information from nearby AVs and intelligent infrastructure to enhance the perception capability. Based on the collaboration strategies, there are three major classes: early [21], late [22], [23], [24], and intermediate fusion [5], [7], [10], [4], [25], [26], [27]. The early fusion method delivers the raw point clouds across agents, and each agent will feed the aggregated point clouds to the network for 3D detection.…”

“…[5] proposes a spatial-aware graph neural network for joint perception and prediction, and [4] employs knowledge distillation to advance the learning with the supervision of early fusion. [7] proposes a location-wise self-attention mechanism to fuse the features from different AVs. This work evaluates all three fusion strategies and the single-agent perception method.…”

“…Thus instead, we design an efficient search algorithm. We adopt the intermediate fusion model AttFuse [7] to construct the collaboration graph where the learnable edge weights are considered to represent the importance of that agent's feature contribution to the ego agent. We leverage these learnable edge weights to define each agent's weakness level so that lower values represent smaller contributions to the overall perception system and thus, assigning these weak agents with higher sampling probability can increase the chance of finding adversarial collaborators with inferior performance.…”

“…It can be sliced and rearranged to get the updated feature H i . Before applying this self-attention model in our scene generator, we will train it on the augmented OPV2V dataset [7] to learn the collaboration graph construction by sending H i to a detection header to produce bounding box predictions. During the adversarial procedure, we fix the network's weights and directly exploit calculated a mn to detect the adversarial collaborators.…”

“…Sooner or later, these autonomous systems will be deployed on roads at scale, opening up opportunities for cooperation between them. Previous works in [4], [5], [6], [7], [8], [9], [10], [11], [12] have demonstrated that by leveraging the Vehicle-to-Everything (V2X) communication technology, AVs and infrastructure can perform cooperative perception by using the shared sensing information and thus significantly enhance the perception performance. Despite the remarkable improvement, these works evaluate the proposed systems on the dataset with natural scenarios that do not contain sufficient safety-critical scenes.…”