Project structure of: google-research/google-research

• google-research Video dynamics analysis and adaptable video timeline modeling.
  • video_structure Analyzing video dynamics with TensorFlow.
    • __init__.py Google-Research/Video_Structure license notice.
    • datasets.py Loads sequence datasets for TensorFlow pipelines.
    • datasets_test.py Test Keras image classification model with data determinism check.
    • dynamics.py TensorFlow VRNN model for video keypoint prediction
    • dynamics_test.py Tests VRNN training, KLDivergence annealing, and dynamics model's sampling schedule.
    • hyperparameters.py Hyperparameters class for video model configuration.
    • losses.py Calculates video trajectory separation loss using Gaussian functions and matrix operations.
    • losses_test.py Tests video structure losses for parallel keypoints.
    • ops.py Video operations for TensorFlow model: keypoint extraction, map scaling, data augmentation.
    • ops_test.py Tests add_coord_channels and heat map functions.
    • README.md Unsupervised learning of object structure and dynamics from videos.
    • requirements.txt TensorFlow GPU version constraint: 1.15.3 or below 2.0.0
    • run.sh Script for setting up and testing video_structure package.
    • train.py Trains video model, predicts keypoints, recons images, compiles and validates.
    • vision.py Vision-enhanced video analysis with encoder-decoder model.
    • vision_test.py Video Autoencoder Model Test with TensorFlow
  • video_timeline_modeling Video Timeline Modeling: Adaptable Components
    • README.md Video timeline modeling dataset for news stories available on Google Drive
    • requirements.txt Dependencies: TensorFlow, PyTorch for video timeline modeling
    • run.sh Bash script sets up Python env, installs deps, runs model.
    • setup.py vtm setup script: Python module installation, no external packages required
    • vtm Video Timeline Modeling with Adaptable Components
      • __init__.py Video timeline modeling license notice
      • dataset.py VTM PyTorch dataset: variable-length video timeline modeling.
      • eval.py Video timeline modeling evaluation script with accuracy and metrics.
      • main.py Video timeline model training with Transformers and distillation.
      • model Video Timeline Modeling with Adjustable Components
        • __init__.py Video Timeline Modeling components import.
        • attention_head.py Video timeline modeling Attention Head class. Calculates attention scores, linear layers, log softmax activation.
        • encoder.py PyTorch Transformer Encoder with Positional Encoding
        • model.py Video Timeline Modeling with Adjustable Classifiers
      • test.py Video timeline modeling module testing.