DeepStream Without SAHI — Crowd Detection with Slice-Trained GELAN-C (448x448 Input)
Автор: Levi Pereira - AI
Загружено: 2026-03-18
Просмотров: 3
Описание:
Baseline aerial detection on a dense crowd scene using NVIDIA DeepStream without SAHI.
The GELAN-C model trained on cropped VisDrone slices (448x448 input) runs on the full frame without slicing, detecting only ~2 objects per frame.
Enabling SAHI boosts detections by over 3600%, as the model needs sliced input at inference to match its training conditions.
Source: • Drone sobrevoa a Avenida Paulista em plena...
Повторяем попытку...
Доступные форматы для скачивания:
Скачать видео
-
Информация по загрузке: