Initialize SAM2 video tracker Args: checkpoint_path: Path to SAM2 checkpoint config_file: Path to SA

Initialize SAM2 video tracker Args: checkpoint_path: Path to SAM2 checkpoint config_file: Path to SA

Initialize SAM controller Args: SAM_checkpoint: Path to SAM2 checkpoint model_type: SAM2 model confi

Initialize SAM2 segmenter Args: SAM_checkpoint: Path to SAM2 checkpoint model_type: SAM2 model confi

Returns the number of videos to generate in one epoch.

r""" Returns: `dict` of attention processors: A dictionary containing all attention processors used in the model with

Converts all parts of a mask to simplified polygons, preserving all disconnected components. The level of simplification is controlled by

Clear all clicks and reset to original first frame

Sets the attention processor to use [feed forward chunking](https://huggingface.co/blog/reformer#2-chunked-feed-forward-layers).

Export a list of frames to a GIF. Args: - frames (list): List of frames (as numpy arrays or PIL Image objects). - output_gif_path (s

Generate mask from clicks on first frame Args: image: np.ndarray, (H, W, 3), RGB image points: np.nd

r""" The [`UNetSpatioTemporalConditionModel`] forward method. Args: sample (`torch.Tensor`): The noisy in

Generates a video by layering foregrounds over a background.

Get mask for first frame only (for preview) Args: frame: np.ndarray, (H, W, 3), uint8 RGB frame poin

Convert click input to prompt format Args: click_state: [[points], [labels]] click_input: JSON string "[[x, y, label]]"

Load SAM2 video tracker with pretrained weights Args: checkpoint_path: Path to SAM2 checkpoint (if None, uses default location)

Load video and extract first frame for mask generation

Load VideoMaMa pipeline with pretrained weights Args: base_model_path: Path to SVD base model (if None, uses default) un

Propagate first frame mask through entire video using SAM2

Draws samples from an lognormal distribution.

Reset the current image

Reset all states

Run SAM2 propagation and VideoMaMa inference together

Add click and update mask on first frame Args: video_state: Dictionary with video data point_prompt: "Positive" or "Nega

Disables custom attention processors and sets the default attention implementation.

Set the current image for segmentation

Track object through video using SAM2 Args: frames: List of numpy arrays, [(H,W,3)]*n, uint8 RGB frames

Run VideoMaMa inference on video frames with mask conditioning Args: pipeline: VideoInferencePipeline instance frames_np