🐛 CameraOrientation.fromDegrees() has LEFT/RIGHT swapped on Android — doesn't round-trip its own .degrees getter

[hmelonjp]

Prerequisites

• I have read and followed every applicable step in the Nitro Modules Troubleshooting guide.
• I have read the VisionCamera Troubleshooting guide.
• I have searched existing issues and found nothing matching.
• I am on the latest version of react-native-vision-camera, or have a specific reason I cannot upgrade.

Reproduction

https://github.com/Melon-Technologies/vc-camera-orientation-fromdegrees-repro

Steps to reproduce

1. Clone https://github.com/Melon-Technologies/vc-camera-orientation-fromdegrees-repro
2. kotlinc camera_orientation_fromdegrees_test.kt -include-runtime -d test.jar
3. kotlin -classpath test.jar Camera_orientation_fromdegrees_testKt
4. Observe: LEFT.degrees=270 -> fromDegrees(270)=RIGHT and RIGHT.degrees=90 -> fromDegrees(90)=LEFT

What did you expect to happen?

CameraOrientation.Companion.fromDegrees(degrees) should be the inverse of CameraOrientation.degrees — i.e. fromDegrees(o.degrees) == o for every CameraOrientation value.

What actually happened?

It only round-trips for UP and DOWN. For LEFT and RIGHT it returns the wrong value:

val CameraOrientation.degrees: Int
  get() = when (this) {
    UP -> 0; DOWN -> 180; LEFT -> 270; RIGHT -> 90
  }

fun CameraOrientation.Companion.fromDegrees(degrees: Int): CameraOrientation {
  val normalizedDegrees = normalizeDegrees(degrees)
  return when (normalizedDegrees) {
    in 45..135 -> CameraOrientation.LEFT    // should be RIGHT (RIGHT.degrees == 90, inside this bucket)
    in 135..225 -> CameraOrientation.DOWN
    in 225..315 -> CameraOrientation.RIGHT  // should be LEFT (LEFT.degrees == 270, inside this bucket)
    else -> CameraOrientation.UP
  }
}

fromDegrees(90) returns LEFT instead of RIGHT; fromDegrees(270) returns RIGHT instead of LEFT. The LEFT/RIGHT buckets are swapped relative to .degrees' own mapping.

Why it matters: ImageProxy.orientation calls fromDegrees(imageInfo.rotationDegrees) to label every incoming camera frame:

val ImageProxy.orientation: CameraOrientation
  get() {
    val degrees = imageInfo.rotationDegrees
    return CameraOrientation.fromDegrees(degrees)
  }

A front-facing sensor held in portrait commonly reports rotationDegrees = 270. Pre-fix, fromDegrees(270) returns RIGHT instead of the self-consistent LEFT — a 180-degree-equivalent mislabeling. That wrong orientation value feeds react-native-vision-camera-resizer's rotationDegrees push constant downstream.

Observed live: on a real Android device (Samsung Galaxy A54 5G, Android 15), a detected face's bounding box was drawn over an unrelated part of the frame instead of the actual face, while the identical JS/native pipeline on iOS (iPhone 12, iOS 16.2) drew it correctly — which is what pointed at this Android-only orientation mapping rather than the detector model itself.

Fix

Swap the LEFT/RIGHT buckets to match .degrees:

 fun CameraOrientation.Companion.fromDegrees(degrees: Int): CameraOrientation {
   val normalizedDegrees = normalizeDegrees(degrees)
   return when (normalizedDegrees) {
-    in 45..135 -> CameraOrientation.LEFT
+    in 45..135 -> CameraOrientation.RIGHT
     in 135..225 -> CameraOrientation.DOWN
-    in 225..315 -> CameraOrientation.RIGHT
+    in 225..315 -> CameraOrientation.LEFT
     else -> CameraOrientation.UP
   }
 }

We've been running this as a local patch-package patch to react-native-vision-camera@5.1.1 in production.

Affected platforms

Android (device)

Device(s) affected

Observed on Samsung Galaxy A54 5G (Android 15). Root cause confirmed by direct source inspection and a pure JVM round-trip test — the bug is deterministic enum-mapping logic independent of specific hardware.

VisionCamera version

5.1.1

React Native version

0.86

React Native architecture

New Architecture (Fabric / bridgeless)

Features being used

• Preview
• Photo capture
• Video capture
• Frame Processors (worklets)
• Skia Frame Processors
• Code/Barcode Scanner
• Location metadata
• Multi-cam
• Depth data
• HDR / custom dynamic range
• Custom format / FPS / resolution

Relevant logs / stack trace

This is a logic/correctness bug, not a crash — there is no stack trace. See the reproduction repo's console output:


CameraOrientation.fromDegrees round-trip test (issue: LEFT/RIGHT swapped)

BUGGY fromDegrees (pristine upstream code):
  UP.degrees=0 -> fromDegreesBuggy(0)=UP  OK
  DOWN.degrees=180 -> fromDegreesBuggy(180)=DOWN  OK
  LEFT.degrees=270 -> fromDegreesBuggy(270)=RIGHT  MISMATCH (bug reproduced)
  RIGHT.degrees=90 -> fromDegreesBuggy(90)=LEFT  MISMATCH (bug reproduced)
  -> all orientations round-trip correctly: NO -- bug confirmed

FIXED fromDegrees (LEFT/RIGHT buckets swapped):
  UP.degrees=0 -> fromDegreesFixed(0)=UP  OK
  DOWN.degrees=180 -> fromDegreesFixed(180)=DOWN  OK
  LEFT.degrees=270 -> fromDegreesFixed(270)=LEFT  OK
  RIGHT.degrees=90 -> fromDegreesFixed(90)=RIGHT  OK
  -> all orientations round-trip correctly: yes, bug is fixed

RESULT: PASS -- bug reproduced on unpatched code, confirmed fixed on patched code.

Additional context

No response

Submission

• The reproduction I linked is either (preferred) a PR against this repo that adds a failing harness test following the harness-tests README, or (fallback) a public repo that reproduces the bug on a fresh clone. I understand the issue will be closed without one.
• I pasted logs as text (not screenshots).
• I wrote this report in my own words. I did not paste AI-generated descriptions of the bug.

[mrousavy]

Thanks for the report! Will look into it

[hmelonjp]

Can we have an update ?

Happy to open a PR with this fix (the LEFT/RIGHT bucket swap in fromDegrees, as described above).

One note on testing: This bug lives in CameraOrientation.fromDegrees(), which is only reachable through the device (accelerometer) orientation source (HybridDeviceOrientationManager/useOrientation('device')). We looked at adding a regression test under apps/simple-camera/tests per your Harness conventions, but since that path is driven by real sensor degrees at runtime and there's no public API to simulate or force a device orientation reading in CI. So we can't add a Harness test that reliably reproduces this.

What we can add is a plain JVM unit test for CameraOrientation.degrees/fromDegrees() . But there's currently no src/test source set or test dependencies configured for the Android module, so this would mean introducing a small new test setup alongside the fix. Would that be acceptable, or would you prefer we submit the fix without a test given the Harness-infrastructure gap?

[mrousavy]

I'll be tackling orientation rewrites soon, I have a better idea