This small project provides a foundation on which to build object detection models in Diablo II: Resurrected. It uses PyTorch as a backend; the current model uses FasterRCNN with a mobilenet_v2 backbone. The animation shown depicts the model performance after 10 epochs of training which required ca. 5 minutes of training time (NVIDIA GTX 1080 Ti).
d2torch is written in Python. We prefer to use Anaconda (specifically miniconda) for managing the build environment. You will also need:
From the cloned repository:
conda env create -f environment.yml
conda activate d2torch
python .\src\train.py
With Diablo II: Resurrected running (in windowed mode):
python .\src\infer.py