Independent project
Cricket-Shot Predictor
Video classification for cricket shots, from frame extraction to a runnable prediction demo.
Built embedding-based video classifiers and fine-tuned VideoMAE.
Approach
Built a video-processing pipeline covering frame extraction, embeddings, labeling, experiment logging and model sharing on Hugging Face.
Explored CLIP and SigLIP embeddings with LSTM classifiers, then fine-tuned VideoMAE.
Results and interpretation
Recorded test accuracy: CLIP–LSTM 52%, SigLIP–LSTM 53%, and fine-tuned VideoMAE 66.05%. Dataset and split documentation is pending, so these results should not be treated as a controlled model comparison.
Try it
The demo and source repository are linked below. They provide a direct way to inspect the implementation and its predictions.
Tools
LSTM · CLIP · VideoMAE · HuggingFace