N00bML/ucf50-action-recognition

🎬 Video Action Recognition

Recognizes human actions in videos using deep learning.

Model: GRU (Gated Recurrent Unit) Accuracy: 97.23% Top-1 | 99.85% Top-5 Dataset: UCF-50 (50 action classes)

How to Use

1. 1.Upload a video (MP4, AVI, MOV)
2. 2.Click "Recognize Action"
3. 3.View predictions with confidence scores!

Supported Actions

50 human activities including:

• —Sports (Basketball, Baseball, Tennis, etc.)
• —Music (Guitar, Piano, Violin, etc.)
• —Fitness (Push-ups, Pull-ups, etc.)
• —Daily activities (Walking, Mixing, etc.)

Model Architecture

• —Feature Extractor: ResNet50 (pretrained)
• —Sequence Model: GRU
• —Input: 32 frames per video
• —Parameters: 3.96M

GitHub

Full Project Repository

Contact

Mushtaq Saeed

• —Email: mushtaqsaeed577@gmail.com
• —GitHub: @NoobML