Getting totally different output from Mediapipe Hands Python API and tflite model

[Mohit0928]

trafficstars

I'm running inference using the Python API of Hand Landmark Detection to generate hand keypoint.

I'm running inference on the same image with tflite model of Hand Landmark Detection.

I haver created a Colab Notebook here where I have used both the approach.

But I'm getting totally different output from both. What's going under the hood? Could anyone please explain why I'm getting different output from both. These should be a bit similar at least.