从图像对的两个图像中分别采样 region 对。

从 image A 采样 object、从 image B 采样 region。

Sample 1 random point (along with its label) from the center of each error region, that is, the point with the largest distance to the bounda

Sample k index pairs with covisibility above covis_threshold and both valid, without repeating images. Optional motion filters (if frames provide

Sample `num_pt` random points (along with their labels) independently from the error regions. Inputs: - gt_masks: [B, 1, H_im, W_im] mas

Save messages to debug jsonl file if debug is enabled

保存点云为 PLY 格式 Args: points: 点云坐标 (N, 3) colors: 点云颜色 (N, 3) filename: 输出文件名

Select up to `max_cond_frame_num` conditioning frames from `cond_frame_outputs` that are temporally closest to the current frame at `frame_id

Wraps a function and prints the shapes of all tensor inputs. Only prints when a new combination of shapes is seen. Thread-safe. Args

Shutdown the predictor and clear all sessions.

Start a new inference session on an image or a video. Here `resource_path` can be either a path to an image file (for image inference

Args: see documentation of :func:`paste_masks_in_image`.

Convert a Color instance to a hex string. Args: color (Color): Color instance of color. Returns: Co

Expects a tensor of shape Bx4. The coordinates can be in absolute image or normalized coordinates, if the coords are in absolute imag

Learning rate scheduling per step

Return a boolean mask for frames that have existing image files and are not marked bad.

Return a boolean mask for frames that have existing image files and are not marked bad.

使用 vLLM 服务验证标签是否正确。

使用 vLLM 服务验证标签是否正确。

Partition into non-overlapping windows with padding if needed. Args: x (tensor): input tokens with [B, H, W, C]. window_size

Window unpartition into original sequences and removing padding. Args: x (tensor): input tokens with [B * num_windows, window_size, w

Returns the tokenized representation of given input string(s) Parameters ---------- texts : Union[str, List[str]]

Returns the item at the specified index. Args: index (int): The index of the item to return. Returns:

Args: item: int, slice, or a BoolTensor Returns: Boxes: Create a new :class:`Boxes` by indexing. Th

Create a new `Keypoints` by indexing on this `Keypoints`. The following usage are allowed: 1. `new_kpts = kpts[3]`: return

Returns: RotatedBoxes: Create a new :class:`RotatedBoxes` by indexing. The following usage are allowed: 1. `new

Returns: BitMasks: Create a new :class:`BitMasks` by indexing. The following usage are allowed: 1. `new_masks =

Support indexing over the instances and return a `PolygonMasks` object. `item` can be: 1. An integer. It will return an obje

Returns: ROIMasks: Create a new :class:`ROIMasks` by indexing. The following usage are allowed: 1. `new_masks =

Remove a few attributes during pickling, so that this async video loader can be saved and loaded as a part of the model session.

Uses SAM-3 to calculate the image embedding for an image, and then allow repeated, efficient mask prediction given prompts.

SimpleFPN neck a la ViTDet (From detectron2, very lightly adapted) It supports a "dual neck" setting, where we have two ident

Args: kernel_size (Tuple): kernel size of the projection layer. stride (Tuple): stride of the projection layer.

Args: dim (int): Number of input channels. num_heads (int): Number of attention heads. qkv_bias (bool: I

Args: dim (int): Number of input channels. num_heads (int): Number of attention heads in each ViT block.

Initialize the backbone combiner. :param visual: The vision backbone to use :param text: The text encoder to use