EasyComputerVision is a Unity plugin that integrates the brand new Sentis with a code-free interface to offer an easy way to run ML/DL computer vision models inside the Unity Game Engine. It allows you to use Sentis code-free and currently provides an easy method for image classification (with more use-cases to come). Simply create an in-game camera or use your webcam and add an ONNX model to your project to easily run a model. EasyComputerVision is created by Futurist Acoustics 
Purchase EasyComputerVision on the Unity Asset Store
Our aim is to see how many of the Computer Vision model types available at Hugging Face we can implement given the constraints of Sentis:
- Image Classification
- Object Detection
- Image Segmentation
- Depth Estimation
- Image-to-Image
- Unconditional Image Generation
- Video Classification
- Zero-Shot Image Classification
A range of CNNs can be used to classify images from an in-game camera or using your webcam. A majority of CNN models work, including for both greyscale and color inputs. The ONNX model zoo has recently been updated with numerous ONNX models.
✅ ViT-Base-Patch16–224 ✅ GoogLeNet ✅ ResNet-50 ✅ ConvNeXt Nano ✅ ConvNeXt Tiny ✅ ConvNeXt Large
- Object detection using YOLO Tiny v7
- Uses the new Unity Sentis, replacing Unity Barracuda
- Show results in console
- Apply softmax to the output to view the probabilities as a percentage
- Input can be normalized. The input image can be set to values between 0 and 1 or 0 and 255
- Can view a summary of the model for convenience, including input and output.
- Scheduled execution can be fine-tuned. The number of model layers processed per frame can be configured, enabling greated control of the inference overhead.
- Image classification
- Can use in-game camera as input
- Can use webcam as input
- Automatically creates a render texture using the model's input dimensions
- Inference on update
- Inference on interval
- Inference on interval using scheduled execution
- Inference on key press
- Inference on key press using scheduled execution
- Can use ComputePrecompiled (GPU - highly efficient GPU code with all overhead code stripped away and precompiled into the worker)
- Can use Compute (GPU - highly efficient GPU but with some logic overhead)
- Can use ComputeRef (GPU - a less efficient but more stable reference implementation)
- Can use CSharpBurst (CPU - highly efficient, jobified and parallelized CPU code compiled via Burst)
- Can use CSharp (CPU - slightly less efficient CPU code)
- Can use CSharpRef (CPU - a less efficient but more stable reference implementation)
- Can display input render texture as image on a canvas
- Can change position and size of canvas image
EasyComputerVision is created by Futurist Acoustics