app.py

import gradio as gr
from datasets import load_dataset as load_dataset_hf, get_dataset_config_names
import os

# Predefined dataset names (configs will be fetched dynamically)
PREDEFINED_DATASETS = [
    "abraranwar/agibotworld_alpha_rfm",
    "abraranwar/libero_rfm",
    "abraranwar/usc_koch_rewind_rfm",
    "aliangdw/metaworld",
    "anqil/rh20t_rfm",
    "anqil/rh20t_subset_rfm",
    "jesbu1/auto_eval_rfm",
    "jesbu1/egodex_rfm",
    "jesbu1/epic_rfm",
    "jesbu1/fino_net_rfm",
    "jesbu1/failsafe_rfm",
    "jesbu1/hand_paired_rfm",
    "jesbu1/galaxea_rfm",
    "jesbu1/mit_franka_p-rank_rfm",
    "jesbu1/utd_so101_clean_policy_ranking_top",
    "jesbu1/utd_so101_clean_policy_ranking_wrist",
    "jesbu1/h2r_rfm",
    "jesbu1/humanoid_everyday_rfm",
    "jesbu1/molmoact_rfm",
    "jesbu1/motif_rfm",
    "jesbu1/oxe_rfm",
    "jesbu1/oxe_rfm_eval",
    "jesbu1/ph2d_rfm",
    "jesbu1/racer_rfm",
    "jesbu1/roboarena_0825_rfm",
    "jesbu1/soar_rfm",
    "jesbu1/usc_koch_human_robot_paired",
    "jesbu1/usc_koch_p_ranking_rfm",
    "ykorkmaz/libero_failure_rfm",
    "aliangdw/usc_xarm_policy_ranking",
    "aliangdw/usc_franka_policy_ranking",
    "aliangdw/utd_so101_policy_ranking",
    "aliangdw/utd_so101_human"
]

def load_rfm_dataset(dataset_name, config_name):
    """Load the RFM dataset from HuggingFace Hub."""
    try:
        # Validate inputs
        if not dataset_name or not config_name:
            return None, "❌ Please provide both dataset name and configuration"
        
        # Try to load the dataset
        dataset = load_dataset_hf(dataset_name, name=config_name, split="train")
        
        # Check if dataset has the expected structure
        expected_features = ["task", "frames", "quality_label", "is_robot", "data_source"]
        missing_features = [f for f in expected_features if f not in dataset.features]
        
        if missing_features:
            return None, f"⚠️ Dataset loaded but missing expected features: {missing_features}"
        
        # Check if dataset has any samples
        if len(dataset) == 0:
            return None, f"⚠️ Dataset {dataset_name}/{config_name} is empty"
        
        return dataset, f"✅ Loaded {len(dataset)} trajectories from {dataset_name}/{config_name}"
    except Exception as e:
        error_msg = str(e)
        if "not found" in error_msg.lower():
            return None, f"❌ Dataset or configuration not found: {dataset_name}/{config_name}"
        elif "authentication" in error_msg.lower():
            return None, f"❌ Authentication required for {dataset_name}"
        else:
            return None, f"❌ Error loading dataset: {error_msg}"

def get_available_configs(dataset_name):
    """Get available configurations for a dataset."""
    try:
        # Use the dedicated function to get config names
        configs = get_dataset_config_names(dataset_name)
        return configs
    except Exception as e:
        print(f"Error getting configs for {dataset_name}: {e}")
        return []

def update_config_choices_with_custom(dataset_name):
    """Update config choices by fetching from the dataset."""
    if not dataset_name:
        return gr.update(choices=[], value="")
    
    try:
        # Always try to fetch configs from the dataset
        configs = get_available_configs(dataset_name)
        if configs:
            current_value = configs[0]
            return gr.update(choices=configs, value=current_value)
        else:
            return gr.update(choices=[], value="")
    except Exception as e:
        # If fetching fails, allow custom input
        print(f"Warning: Could not fetch configs for {dataset_name}: {e}")
        return gr.update(choices=[], value="")

def visualize_trajectory(dataset, index, dataset_name=None):
    """
    Function to retrieve a trajectory and its metadata from the dataset.
    """
    if dataset is None:
        return None, "Error: Could not load dataset", "Error: Could not load dataset", None
    
    try:
        item = dataset[int(index)]
        
        # Get metadata
        task = item["task"]
        quality_label = item["quality_label"]
        is_robot = item["is_robot"]
        data_source = item["data_source"]
        
        # Get the frames data (video file path)
        frames_data = item["frames"]
        
        # Handle video file path (could be local path or HuggingFace Hub URL)
        if isinstance(frames_data, str):
            # Use dynamic dataset name if provided, otherwise fallback to default
            if dataset_name:
                video_path = f"https://huggingface.co/datasets/{dataset_name}/resolve/main/{frames_data}"
            else:
                video_path = f"https://huggingface.co/datasets/aliangdw/rfm/resolve/main/{frames_data}"
            frames_info = f"Video file: {video_path}"
        else:
            return None, "Error: Invalid video path", "Error: Invalid video path", None
        
        # Create metadata
        metadata = f"""
        ## Trajectory Information

        **Video path:** {video_path}
        
        **Language Task:** {task}
        
        **Quality Label:** {quality_label}
        
        **Data Type:** {'Robot' if is_robot else 'Human'}
        
        **Source:** {data_source}
                
        **Trajectory ID:** {item.get('id', 'N/A')}
        """
        
        # Return video path for Gradio to display
        return video_path, metadata, f"Trajectory {index}", None
        
    except Exception as e:
        return None, f"Error: {str(e)}", f"Error: {str(e)}", None

# Create the Gradio interface
with gr.Blocks(title="RFM Dataset Visualizer") as demo:
    gr.Markdown("# RFM Dataset Visualizer")
    gr.Markdown("Browse through trajectory videos and their metadata from the Reward Foundation Model dataset.")
    
    # Dataset selection
    with gr.Row():
        with gr.Column(scale=2):
            dataset_name_input = gr.Dropdown(
                choices=PREDEFINED_DATASETS,
                value="jesbu1/oxe_rfm",
                label="Dataset Name",
                allow_custom_value=True
            )
        
        with gr.Column(scale=2):
            config_name_input = gr.Dropdown(
                choices=[],
                value="",
                label="Configuration Name",
                allow_custom_value=True
            )
        
        with gr.Column(scale=1):
            refresh_btn = gr.Button("🔄 Refresh Configs", variant="secondary", size="sm")
        
        with gr.Column(scale=1):
            load_btn = gr.Button("Load Dataset", variant="primary")
    
    # Status message
    status_output = gr.Markdown("Ready to load dataset...")
    
    # Dataset info
    dataset_info = gr.Markdown("")
    
    # Visualization section
    with gr.Row():
        with gr.Column(scale=2):
            # Video/Image display
            video_output = gr.Video(label="Trajectory Video", height=400, autoplay=True)
            image_output = gr.Image(label="Frame Preview", height=400, visible=False)
        
        with gr.Column(scale=1):
            # Metadata display
            metadata_output = gr.Markdown(label="Metadata")
    
    # Navigation controls
    with gr.Row():
        with gr.Column(scale=1):
            prev_btn = gr.Button("⬅️ Previous", variant="secondary")
        
        with gr.Column(scale=2):
            # Slider for navigation with dynamic max
            slider = gr.Slider(
                minimum=0,
                maximum=0,
                step=1,
                value=0,
                label="Select a dataset first",
                interactive=False
            )
        
        with gr.Column(scale=1):
            next_btn = gr.Button("Next ➡️", variant="secondary")
    
    # Current trajectory title
    title_output = gr.Textbox(label="Current Trajectory", interactive=False)
    
    # State variables
    current_dataset = gr.State(None)
    current_index = gr.State(0)
    
    def load_dataset(dataset_name, config_name):
        """Load the dataset and update the interface."""
        dataset, status = load_rfm_dataset(dataset_name, config_name)
        if dataset is not None:
            max_index = len(dataset) - 1
            info = f"**Dataset Info:**\n- **Total Trajectories:** {len(dataset)}\n- **Features:** {list(dataset.features.keys())}"
            # Return the dataset length for number input configuration
            return dataset, status, info, 0, max_index
        else:
            return None, status, "", 0, 0
    
    def update_trajectory(dataset, index, dataset_name=None):
        """Update the displayed trajectory."""
        if dataset is None:
            return None, "No dataset loaded", "No dataset loaded", None
        # Ensure index is within bounds and is a valid number
        if index is None or not isinstance(index, (int, float)):
            index = 0
        elif index >= len(dataset):
            index = len(dataset) - 1
        elif index < 0:
            index = 0
        return visualize_trajectory(dataset, int(index), dataset_name)
    
    def next_trajectory(dataset, current_idx, dataset_name=None):
        """Go to next trajectory."""
        if dataset is None:
            return current_idx, None, "No dataset loaded", "No dataset loaded", None
        next_idx = min(current_idx + 1, len(dataset) - 1)
        video, metadata, title, image = visualize_trajectory(dataset, next_idx, dataset_name)
        return next_idx, video, metadata, title, image
    
    def prev_trajectory(dataset, current_idx, dataset_name=None):
        """Go to previous trajectory."""
        if dataset is None:
            return current_idx, None, "No dataset loaded", "No dataset loaded", None
        prev_idx = max(current_idx - 1, 0)
        video, metadata, title, image = visualize_trajectory(dataset, prev_idx, dataset_name)
        return prev_idx, video, metadata, title, image
    
    def update_slider_range(dataset):
        """Update the slider with new maximum value based on dataset length."""
        if dataset is not None:
            max_value = len(dataset) - 1
            return gr.update(
                maximum=max_value,
                value=0,  # Reset to beginning
                label=f"Trajectory Index (0 to {max_value})",
                interactive=True
            )
        else:
            return gr.update(
                maximum=0,
                value=0,
                label="Select a dataset first",
                interactive=False
            )
    
    def update_config_choices(dataset_name):
        """Update the config dropdown choices based on selected dataset."""
        return update_config_choices_with_custom(dataset_name)
    
    # Connect the components
    # Update config choices when dataset changes
    dataset_name_input.change(
        fn=update_config_choices,
        inputs=[dataset_name_input],
        outputs=[config_name_input]
    )
    
    # Refresh configs button for custom datasets
    refresh_btn.click(
        fn=update_config_choices_with_custom,
        inputs=[dataset_name_input],
        outputs=[config_name_input]
    )
    
    load_btn.click(
        fn=load_dataset,
        inputs=[dataset_name_input, config_name_input],
        outputs=[current_dataset, status_output, dataset_info, current_index]
    ).then(
        fn=update_slider_range,
        inputs=current_dataset,
        outputs=slider
    )
    
    slider.change(
        fn=lambda dataset, idx, dataset_name: update_trajectory(dataset, idx, dataset_name),
        inputs=[current_dataset, slider, dataset_name_input],
        outputs=[video_output, metadata_output, title_output, image_output]
    ).then(
        fn=lambda dataset, idx: idx if dataset is None or idx < len(dataset) else len(dataset) - 1,
        inputs=[current_dataset, slider],
        outputs=[current_index]
    )
    
    next_btn.click(
        fn=lambda dataset, idx, dataset_name: next_trajectory(dataset, idx, dataset_name),
        inputs=[current_dataset, current_index, dataset_name_input],
        outputs=[current_index, video_output, metadata_output, title_output, image_output]
    ).then(
        fn=lambda idx: idx,
        inputs=current_index,
        outputs=slider
    )
    
    prev_btn.click(
        fn=lambda dataset, idx, dataset_name: prev_trajectory(dataset, idx, dataset_name),
        inputs=[current_dataset, current_index, dataset_name_input],
        outputs=[current_index, video_output, metadata_output, title_output, image_output]
    ).then(
        fn=lambda idx: idx,
        inputs=current_index,
        outputs=slider
    )
    
    # Load initial dataset and configs
    demo.load(
        fn=lambda: ("jesbu1/oxe_rfm", "oxe_jaco_play"),  # Set initial values
        outputs=[dataset_name_input, config_name_input]
    ).then(
        fn=update_config_choices_with_custom,
        inputs=[dataset_name_input],
        outputs=[config_name_input]
    ).then(
        fn=lambda dataset_name, config_name: load_dataset(dataset_name, config_name),
        inputs=[dataset_name_input, config_name_input],
        outputs=[current_dataset, status_output, dataset_info, current_index]
    ).then(
        fn=update_slider_range,
        inputs=current_dataset,
        outputs=slider
    )

def main():
    """Main function to launch the RFM visualizer."""
    demo.launch()

# Launch the app
if __name__ == "__main__":
    main()