hthuwal/sign-language-gesture-recognition — reverse-engineered prompt

Build me a Python project that can recognize sign language gestures from short video clips. I want to be able to put training and test videos into class folders, extract frames from each video, keep the simple hand segmentation step if it helps, retrain an image model on those frames, then turn each video into a sequence of frame based features or class probabilities and train an RNN to predict the final gesture for the whole clip.

Please make the whole flow easy to run from the command line, from raw videos to saved model checkpoints and test results. It should work with the Argentinian Sign Language dataset style shown in the README, but also be okay for a small dummy dataset with just a few classes. I would like clear outputs for labels, saved models, and prediction results, plus a basic way to inspect what the model guessed on test videos. Please clean up anything brittle, keep the code understandable, and update the README so a beginner can follow it step by step. Look up current docs online if you need to.