Bad scaling rendering ARKit

I'm working on perspective tests with ARKit and SceneKit. The idea is to improve 3D rendering when displaying a flat 3D model on the ground. I had already opened a ticket to another perspective problem that is almost solved. (https://stackoverflow.com/questions/51377892/arkit-perspective-rendering?noredirect=1#comment89803924_51377892)

However, I noticed after my multitudes tests / 3D display, that sometimes when I anchor a 3D model, the size of it can differ... (width and length)

I usually display a 3D model that is 16 meters long and 1.5 meters wide. You can well imagine that this distorts my rendering.

I don't know why my display may differ in terms of 3D model size.

Maybe it's from the tracking and my test environment.

Below is the code I use to add my 3D model to the scene:

   func renderer(_ renderer: SCNSceneRenderer, didAdd node: SCNNode, for anchor: ARAnchor) {
        guard let imageAnchor = anchor as? ARImageAnchor else { return }
        let referenceImage = imageAnchor.referenceImage
        let imageAnchorPosition = imageAnchor.transform.columns.3
        print("Image detected")
       
        let modelName = "couloirV2"
        //let modelName = "lamp"
        guard let object = VirtualObject
            .availableObjects
            .filter({ $0.modelName == modelName })
            .first else { fatalError("Cannot get model \(modelName)") }
       
        print("Loading \(object)...")
       
        self.sceneView.prepare([object], completionHandler: { _ in
            self.updateQueue.async {
                // Translate the object's position to the reference node position.
                object.position.x = imageAnchorPosition.x
                object.position.y = imageAnchorPosition.y
                object.position.z = imageAnchorPosition.z
               
                // Save the initial y value for slider handler function
                self.tmpYPosition = object.position.y
               
                // Match y node's orientation
                object.orientation.y = node.orientation.y
                print("Adding object to the scene")
               
                // Prepare the object
                object.load()
               
                // Show origin axis
                object.showObjectOrigin()
               
               
                // Translate on z axis to match perfectly with image detected.
                var translation = matrix_identity_float4x4
                translation.columns.3.z += Float(referenceImage.physicalSize.height / 2)
                object.simdTransform = matrix_multiply(object.simdTransform, translation)
               
                self.sceneView.scene.rootNode.addChildNode(object)
                self.virtualObjectInteraction.selectedObject = object
                self.sceneView.addOrUpdateAnchor(for: object)
            }
        })
    }