Add bicycle inside image detection for new BikeTag Post submissions.

[KenEucker]

Is your feature request related to a problem? Please describe.
Currently, Players of BikeTag can upload any image to the site.

Describe the solution you'd like
We can use ML, specifically ml5.js, to detect whether or not a bicycle is present in the image. If a bicycle is not present, the Player should see a similar popup to when an image is submitted outside of the boundary of a game: "The photo you uploaded does not appear to have a bicycle in it. Please upload a photo with a bicycle clearly visible."

Additionally, since BikeTag Posts are not "selfies", we could detect whether or not a "face" is present in the photo and reject is based on that. The API method that handles this, validateBikeTagImage, should return a boolean for valid and an error object, which is undefined if it is a valid image.

Describe alternatives you've considered
We've talked about this in the past and some solutions were discussed then. Now, after looking into it for a few minutes, I've chosen ml5.js. Other solutions are available and anything that is free will be considered.

Additional context
ML5JS - Object Detector

[KenEucker]

Here's the relevant ML5JS code:

// When the model is loaded
function modelLoaded() {
  console.log('Model Loaded!');
}

// Create a ObjectDetector method
const objectDetector = ml5.objectDetector('cocossd', {}, modelLoaded);

// Detect objects in the image element
objectDetector.detect(image, (err, results) => {
  console.log(results); // Will output bounding boxes of detected objects
});

[KenEucker]

ml5js documentation has moved: https://docs.ml5js.org/#/reference/image-classifier

However, I have tried using this code with a recent BikeTag image and none of the available models were able to find "bicycle" from the image. The test image clearly had a largely visible bicycle in it, in the center of the image, which was slightly obfuscated by landscaping directly behind it. I do not believe that ml5js will work at all for our purposes, unfortunately.

In further testing of BikeTag images, bicycles were found in other photos. This code was about 50% reliable in finding a bicycle in the photo. I still think we need a higher effective rate than 50%.

[KenEucker]

Cursory testing of MediaPipe, from Google, shows better results:

https://ai.google.dev/edge/mediapipe/solutions/vision/object_detector/web_js