Dynamic batch size

[Deepakpy]

The onnx models like hand_landmark_080_Nx3x224x224.onnx under https://github.com/PINTO0309/PINTO_model_zoo/tree/main/033_Hand_Detection_and_Tracking/30_batchN_post-process_marged. Have dynamic batch size. When its converted to Tensorflow or tflite model the batch size is converted to 1[1x3x224x224].
Does Tensorflow support dynamic batch size. If so is it possible to convert the onnx model to tflite retaining the dynamic batch size?
Any advice will be helpful to understand the conversation process more deeply

[PINTO0309]

https://github.com/PINTO0309/onnx2tf

[Deepakpy]

The quantized palm model palm_detection_builtin_256_integer_quant.tflite is having input size of 1x256x256x3 which is same as the size of mediapipe blazepalm model.
The newer mediapipe palm models are having 1x192x192x3
Where as the palm models under 30_batchN_post-process_merged :

palm_detection_1x3x128x128_post.onnx

These models have different input HW. I wanted to know why this H,W were choosen.