ML pipeline use to auto time speedruns

YOLOv8 model, Paperspace workflows

Handmade dataset

Object detection dataset

Classification dataset

In order to run this script, you need might need to install OpenH624 for opencv. Put open264 dll inside AppData\Local\Programs\Python\Python310\Lib\site-packages\cv2 or the location that contains opencv

Install Python at least 3.8 and install the required package. Virtual environment is optional

Windows venv:

py -m venv .venv
.venv\Scripts\activate.bat

Mac or Linux (py for windows and python or python3 for Mac/Linux):

python -m venv .venv
source .venv/bin/activate

Install required packages:

pip install -U -r requirements.txt

Go to Pytorch website and install torch with the corresponding CUDA version. If you are not training using gpu, just install pytorch normally, if you are training it's required to install CUDA Toolkit and cuDNN and follow instructions at pytorch website to install the according pytorch package

No training:

pip install pytorch

Go to https://github.com/cisco/openh264/releases, download the openh264 dll and put into cv2 package folder IF you need standard mp4 format. Installing Openh264 is required to export inferred video in AVC1 codec (mp4 format), or else OpenCV won't export the video. You can just use XVID (avi or mkv format) and not bother with openh264, as long as you don't bother with video codec problem.

Go to https://ffmpeg.org/, download and add ffmpeg binary to environment variables. FFMPEG is not needed with new verion of the script

Create a folder auth and 2 files email.txt and password.txt inside the folder containing login information of your The Golden House account.

Run this to fetch all unverified Abyss speedruns and start timing:

py entry.py

You need an Agent account or an Admin account on TGH to perform this action