Training Models
To train a model, you will need to play your game of choice while leaving Nezur running in the background to learn from your gameplay.
Last updated
To train a model, you will need to play your game of choice while leaving Nezur running in the background to learn from your gameplay.
Last updated
🤖 Models - Guide
What are Models? Nezur works on AI. The AI collects images that will help the Aim Aligner be more accurate. Models can be manually created by users, shared, or modified to improve its capability. Essentially, the more images that are trained of enemy player skins, the more efficient the model configuration will be. We also have an option 'Collect Data While Playing' to automatically train images too.
Terminology:
Imagine Training: In AI, the term "Trained Images" refers to the process a computer has learned from to understand things like recognizing objects or patterns.
Labeling Images: In AI, the term "Labeling images" refers to the process of adding meaningful information to images specifically to train machine learning models.
Creating Models: An ONNX model file is like a ready-to-use cookbook for teaching computers to understand images. It contains all the instructions and knowledge needed for an AI to recognize things in pictures. This file can be easily shared and used by different AI systems, making it convenient for developers to build image-related applications. In the case of Nezur, .onnx files are used for recognizing images that Nezur needs for its Aim Aligner to function.
This guide will explain step by step how to create your own models.
Collecting Images:
Manual Method - Download images online such as skins of enemy players. You can find model packs, screenshot manually from in-game or videos, and find other resources.
Auto Train Method (Fastest) - Use Nezur's 'Collect Data While Playing' feature with 'Aim Only On Trigger Button' enabled. When the Trigger Button is clicked, a screenshot is snapped. These images can be found inside the bin > images folder from Nezur. Tip: When training images - whether it's manual or automated, it's important to get high-quality images, different skins with added cosmetics, and to get as much detail as possible. Such as different positions of the body (Running, jumping, crouching, etc). It's also recommended to snapshot images with different distances and to remove images that are irrelevant. Collect and use images that will help train the Aim Aligner to be more accurate in locking onto enemies.
Labeling Images: Step 1. Visit [] and click 'Get Started'. Step 2. Click 'Drop an image' (The more Collected Images the better). Step 3. Click 'Object detection'. Step 4. Click 'Start project'. Step 5. In the top left corner under 'Actions' click 'Run AI Locally'. Step 6. Select the 'YoloV5' option and click 'Use model!'. Step 7. Click the down arrow and select 'Yolov5n/COCO' then click 'Use model!'. Step 8. Click 'Select all' and Accept. Deselect any model that's not relevant. Step 9. Under 'Actions' click 'Export Annotations' and choose the first option. Step 10. Save the Model onto your desktop for easier access.
Training Images Required: Step 1. - Install []. Step 2. Install Ultralytics:
Run CMD as Administrator.
Copy and paste the below command into CMD to install Ultralytics (Must have Python first): pip install ultralytics. Note: If you load up CMD and get this error 'pip' is not recognized as an internal or external command, operable program, or batch file. Watch the tutorial [] to fix the path error. Step 3. - Install PyTorch (Optional):
This will train images from your GPU instead of CPU, making the process 10X quicker.
To install PyTorch check the following page []. Choose the CUDA option.
To check if PyTorch is installed open Python and run these two commands:
If the result outputs as True then PyTorch is successfully installed and running.
Train Path - Go to images > train > Copy Path. Example - [C:\Users\User\OneDrive\Desktop\Image Training\images\train].
Val Path - Go to images > val > Copy path. Example - [C:\Users\User\OneDrive\Desktop\Image Training\images\val]. Step 3. Upload the collected images inside the images folder under train. Step 4. Upload the labeled images inside the labels folder under train. Step 5. Run the following commands in CMD to train the images.
Train Command:
Replace the 'putyourmodelnamehere' part with the name of the folder. On default, this should be 'Image Training' make sure there aren't any added spaces or that the name is different. Note: If you increase the epochs number it will increase the speed of the model. After running this command- it will start training images in 'runs' folder that it creates.
Export Model Command:
Boom! That's it. You've now created your own model from scratch and ready to use.
Tutorial of Training Images: Step 1. Download []. (The pack contains a ready-made folder structure and the tool required for training images). Step 2. Open the data.yaml file and set the correct paths.