GPU-accelerated Semantic Image Segmentation with PyTorch

Implementasi program untuk webinar BISA AI - Kampus Merdeka:

Step:

1. Download dan extract dataset cityscapes
   • leftImg8bit_trainvaltest.zip (11GB)
   • gtFine_trainvaltest.zip (241MB)
2. Run jupyter notebook dan buka semseg.ipynb
3. Jalankan program sesuai instruksi yang tertulis di notebook
4. Trained model untuk inference dapat didownload di sini

Other:

1. Deep learning juga dapat digunakan untuk memproses multiple input dan output.
   • O. Natan and J. Miura, "Semantic Segmentation and Depth Estimation with RGB and DVS Sensor Fusion for Multi-view Driving Perception," in Proc. Asian Conf. Pattern Recognition (ACPR), Jeju Island, South Korea, Nov. 2021, pp. 352–365. [paper] [code]
2. Proses learning untuk multiple task dapat diseimbangkan dengan algoritma tertentu.
   • O. Natan and J. Miura, “Towards Compact Autonomous Driving Perception with Balanced Learning and Multi-sensor Fusion,” IEEE Trans. Intelligent Transportation Systems, 2022. [paper] [code]
3. Setelah tahapan perception, planning dan control juga dapat dilakukan secara simultan.
   • O. Natan and J. Miura, “End-to-end Autonomous Driving with Semantic Depth Cloud Mapping and Multi-agent,” IEEE Trans. Intelligent Vehicles, 2022. [paper] [code]
   • O. Natan and J. Miura, “DeepIPC: Deeply Integrated Perception and Control for Mobile Robot in Real Environments,” arXiv:2207.09934, 2022. [preprint] [video]