Custom Instance Segmentation

Autonomous Machines Jetson & Embedded Systems Jetson Projects
1

Custom instance segmentation using detectron2:
GitHub: Custom-object-detection-using-Detectron2/Custom_Instance_Segmentation.ipynb at main · jakariaemon/Custom-object-detection-using-Detectron2 · GitHub

Custom Dataset: https://app.roboflow.com/ds/iB0WrTySI1?key=YTXbygp65f

Trained on RTX 3060 and Tested on Jetson Nano (4GB).
Installation of Detectron2 is straight forward.

2 Likes
2

Have you convert torch weights-model into tensorrt or do you know how to do it, because inferencing net from pytorch on jetson means loosing most of computational performance of the device

4

Hi andriizelenko11, this might be useful for your question of converting to TensorRT: GitHub - NVIDIA-AI-IOT/torch2trt: An easy to use PyTorch to TensorRT converter