Enhancing ADAS Object Detection with YOLOv7: Overcoming Data Shifts through Latent Space Analysis and Efficient Data Synthesis
Tensorleap enables efficient exploration of a model’s latent space to detect data shifts in unseen samples, including unlabeled data. This repository showcases how to detect and address such issues, specifically within an object detection task for ADAS.
Domain Gap
This project addresses a domain gap between KITTI and Cityscapes datasets for semantic segmentation in autonomous driving scenarios. By applying domain adaptation techniques and designing a robust model, we aim to improve cross-dataset generalization. The project seeks to enhance the model’s real-world applicability and contribute to advancements in domain adaptation research.
CityScapes – Object Detection
This project utilizes the YOLOv7 model to perform object detection task on Cityscapes dataset. The cityscapes dataset is a large-scale dataset that stands as one of the standard advanced driver-assistance system (ADAS) benchmarks for multiple vision-related tasks.
Optical Flow – Raft
This project utilizes the RAFT (Recurrent All-pairs Field Transforms) model that estimates Optical Flow using the KITTI dataset.