There are the two models used in Low-light Dataset Project.

The results of the dataset project are uploaded on the AI-Hub

Each folder describe their code, pretrained model and results.

The final low-light result dataset consists of 2 million images.

Train : Valid : Test = 8 : 1 : 1

Object Detection part used Scale-Aware Trident Networks for Object Detection paper.

The original code is from TridentNet detectron2 Github

Instance Segmentation part used PointRend: Image Segmentation as Rendering paper.

The original code is from PointRend detectron2 Github

This code is implemented with Python 3.6 (Anaconda)

Python == 3.6
CUDA >= 11.1
torch == 1.11.0
opencv-python
scikit-image
detectron2