Video Diffusion Project

Baseline Model

• Model: Google Video Diffusion Model
• Backbone: U-Net
• Framework: PyTorch

Datasets

Example

- **Dataset Name**: UCF101
  - **Description**: 101 action classes
  - **Size**: 13,320 videos
  - **Format**: AVI
  - **Source**: https://www.crcv.ucf.edu/data/UCF101.php
  - **License**: MIT

Example

- **Dataset Name**: Kinetics 600
  - **Description**:
  - **Size**:  videos
  - **Format**: 
  - **Source**: https://opendatalab.com/OpenMMLab/Kinetics600/
  - **License**: 

Example

- **Dataset Name**: BAIR_robot_pushing_dataset_v0
  - **Description**:
  - **Size**:  videos
  - **Format**: 
  - **Source**: http://www.google.com/url?q=http%3A%2F%2Frail.eecs.berkeley.edu%2Fdatasets%2Fbair_robot_pushing_dataset_v0.tar&sa=D&sntz=1&usg=AOvVaw3nXoeXKb5FbfJZIUaMX0p_
  - **License**: 

Example

- **Dataset Name**: Kinetics 600
  - **Description**:
  - **Size**:  videos
  - **Format**: 
  - **Source**: https://opendatalab.com/OpenMMLab/Kinetics600/
  - **License**: 

Training Configuration

• Batch Size: 1
• Epochs: 100
• Learning Rate: 1e-4
• Weight Decay: 1e-6
• Timesteps: 1000
• Objective: pred_x0

Evaluation Metrics

• FVD (Fréchet Video Distance)
• FID (Fréchet Inception Distance)

Usage

# Train model
python main.py

# Evaluate model
python evaluate.py