EasyOpenCV VS EasyTensorflowAPI

you can record from the robot's POV with openCV: https://github.com/OpenFTC/EasyOpenCV/blob/master/examples/src/main/java/org/firstinspires/ftc/teamcode/PipelineRecordingExample.java

The SDK provides a UVC driver for webcams; it does not provide anything for phone cameras. If you'd really like to write your own implementation, you can use parts of https://github.com/OpenFTC/EasyOpenCV/blob/master/easyopencv/src/main/java/org/openftc/easyopencv/OpenCvWebcamImpl.java as a reference (which uses the SDK UVC driver), but do note that writing your own implementation means you need to write your own code for a live view, as well as handle threading etc.

getting basic computer vision working is pretty easy with EasyOpenCV and the AprilTag extension.

Apriltags, yes. You just upload the aar files using the OnBot interface. You can get them from the OpenFTC project: https://github.com/OpenFTC/EasyOpenCV/releases/tag/v1.5.2 https://github.com/OpenFTC/EOCV-AprilTag-Plugin/issues/4

Note: https://github.com/OpenFTC/EasyOpenCV/issues/58

We used the example for https://github.com/OpenFTC/EasyOpenCV/releases. Our fork is from the official FTCRobotcontroller repository with extensive slipstreaming. You are welcome to view our repository at baqwas and provide feedback to our programmers who are continuously learning new techniques. I'm sure that they will learn something from your feedback too.

This seems really odd to me, although someone else opened an issue about it: https://github.com/OpenFTC/EasyOpenCV/issues/55

EasyOpenCV

The FTC SDK does not support Image Classification models. If you want to use the model you created with Teachable Machine, you will need something that can use the image classification model. I have not used it, but https://github.com/OutoftheBoxFTC/EasyTensorflowAPI is a library that supports image classification. Its readme says "Image Classification system is only partially tested".

There is a library for using image classification models in the SDK here (https://github.com/OutOfTheBoxFTC/EasyTensorflowAPI), but while it technically has onbot support, it's not recommended due to it's questionable stability on onbot.

So basically, you want to export your Tensorflow model as a .tflite model. This model contains the "brain" of the AI. If the model is a Tensorflow Object Detection model or a Tensorflow Image Classification model you can try to write your own code or use the dev branch of EasyTensorflowAPI, if its a custom one then right now you need to write your own interface using the tensorflow lite API. Basic usage is to put the .tflite file somewhere accessible (EasyTensorflowAPI uses the assets folder of the FTCRobotController module), then load it in an interpreter and run it using that interpreter