christofhenkel / kaggle-asl-fingerspelling-1st-place-solution Goto Github PK

Hi:

Thank you for sharing the code. After training the model, they are saved to:
datamount/weights/cfg_2/fold-1/model.tflite
datamount/weights/cfg_2/fold-1/inference_args.json

But then, how to inference the model? Or, how to use the model to predict a new keypoint sequence? Could you please provide the corresponding code? Thanks!

Hi,
Thanks for your solution!
Can you give an example of using the model in Android Kotlin (or Java)?
For some reason, the model accepts only "one vector" in Kotlin, and outputs "one frame size output"...
The input tensor get by "interpreter.getInputTensor(0)" is fixed 1560 bytes.
And the fixed output get from "interpreter.getOutputTensor(0)" is 4 byets.

How to input x num of frames and get the output correctly in Kotlin/Java?

Thank you!