Camera Capture App on Android

[bearprada]

Here is some things need to discuss

1. Which format the gyro sensor information need to put into EXIF format?

I wrap the x/y/z float value into a JSON body. e.g, {"x":0.12345678, "y":0.12345678, "z":0.12345678}

2. Which EXIF filed?

I can't put the data into the custom tag, so I just use UserComment

3. EXIF tag has size limitation, I can only see 41 characters. (I used this library https://pypi.python.org/pypi/ExifRead)
4. Photo size, do you want super high resolution? e.g, Samsung S7 will have 3024x4032 output size.

ref : result image file : https://www.dropbox.com/s/nwwvzcoxubatag6/android-exif-test.jpg?dl=0

[denkeni]

1. Which format the gyro sensor information need to put into EXIF format?
2. Which EXIF filed?

In previously test, we did use UserComment field of EXIF, and wrapped data into JSON format too, like this:

{
  "lengthInPixel": 200.0,
  "lengthInCentiMeter": 3.79926,
  "plantName": "a"
}

and we maintained another table for plantName.

Currently, we may ignore AR information and focus on viewing angle.

For iOS, I plan to incorporate Euler angles values (unit in radian) [ref]

I had a quick survey about Android (Computing the Device's Orientation and getOrientation). Please check if it's reasonable for Android.

3. EXIF tag has size limitation, I can only see 41 characters. (I used this library https://pypi.python.org/pypi/ExifRead)

I doubt EXIF has such restrictive limit... will check it later.

4. Photo size, do you want super high resolution? e.g, Samsung S7 will have 3024x4032 output size.

We don't really need high resolution images. For iOS, the resolution of image is just the resolution of the screen (just 640×1136 for 4 inch iPhone). There could even be downsampling for AI training.

ref : result image file : https://www.dropbox.com/s/nwwvzcoxubatag6/android-exif-test.jpg?dl=0

Weird that I cannot see UserComment field in your sample image. I'm using Preview on Mac, and there's a Tools > Inspector to view Exif info. Will look into this issue later.

[denkeni]

I suggest suspending EXIF related issues for now, because inference of on-device machine learning models is much more important. That is,

• getOrientation on Android
• inference on TensorFlow Lite

We should focus on these top priorities to make sure they really work.