Face detection not working on version 0.6.0-alpha-1

[cralcubo]

Good afternoon,
Using the pre-trained model UltraFace640 to detect faces on a picture it throws the following exception:

Exception in thread "main" java.lang.IllegalArgumentException: Data shape for input input does not match the model input shape. Data shape: [1, 3, 640, 480], model shape: [1, 3, 480, 640]
	at org.jetbrains.kotlinx.dl.onnx.inference.OnnxInferenceModel.toTensors(OnnxInferenceModel.kt:227)
	at org.jetbrains.kotlinx.dl.onnx.inference.OnnxInferenceModel.predict(OnnxInferenceModel.kt:212)
	at org.jetbrains.kotlinx.dl.onnx.inference.OnnxInferenceModel.predict(OnnxInferenceModel.kt:198)
	at org.jetbrains.kotlinx.dl.onnx.inference.OnnxInferenceModel.predict(OnnxInferenceModel.kt:190)
	at org.jetbrains.kotlinx.dl.onnx.inference.OnnxHighLevelModel$DefaultImpls.predict(OnnxHighLevelModel.kt:50)
	at org.jetbrains.kotlinx.dl.onnx.inference.facealignment.FaceDetectionModelBase.predict(FaceDetectionModelBase.kt:18)
	at org.jetbrains.kotlinx.dl.onnx.inference.facealignment.FaceDetectionModelBase.predict(FaceDetectionModelBase.kt:18)
	at org.jetbrains.kotlinx.dl.onnx.inference.facealignment.FaceDetectionModelBase.detectFaces(FaceDetectionModelBase.kt:44)

The sample code used to reproduce it is as follows:

    val modelHub = ONNXModelHub(cacheDirectory = File("cache/pretrainedModels"))
    val model = modelHub.loadPretrainedModel(ONNXModels.FaceDetection.UltraFace640)
    model.use { faceDetectionModel ->
        val imageFile = getFileFromResource("org/example/$picture")
        val image = ImageConverter.toBufferedImage(imageFile)
        val faces = faceDetectionModel.detectFaces(image, -1)
        visualize(imageFile, faces)
    }

Kind regards and really looking forward to hearing from you 💯

[dosier]

Reading the error, it seems that ur dimensions are swapped, the model expecting 480x640, but u are giving it 640x480 (it's probably rotated 90 degrees?)

[cralcubo]

Thanks for the prompt reply!
What triggered my surprise is that the provided snippet works with Kotlin DL v0.5.2 but it does not with 0.6.0-alpha-1.
Also this problem seems to be related only with face detection models; I tried it with object detection (SSDMobileNetV1) and it brings the expected results.
Kind regards!