Find proper places to draw text given a binary mask.

Find proper places to draw text given a binary mask.

Encode the current image and its prediction into a memory feature.

通过 mask_path 从 mask_index.json 中查找对应的 mask RLE 数据。

fallback: 当 mask_index.json 里缺失 mask_rle 时，从 bbox json 的 instance.mask_encodings 获取。 对齐规则：优先 images 与 mask_encodings 同长度且按 index 对齐；否则尝试 base

Forward SAM prompt encoders and mask heads. Inputs: - backbone_features: image features of [B, C, H, W] shape - poin

从 RLE mask 计算 2D 边界框 (x_min, y_min, x_max, y_max)。

从 RLE mask 计算 2D 中心点。

Get connected components from masks (possibly padding them to an even size).

Retrieve correct image features from backbone output.

Get the Tracker inference states that contain the given object ids. This is used to run partial Tracker propagation on a single objec

A context which ignores CUDA OOM exception from pytorch.

Initialize metadata for the masklets.

从 images 路径列表中推断并加载 mask_index.json 文件。 Args: scene_id: 场景 ID images: mask 图像路径列表（来自 bbox_json 的 images 字段） Returns:

从 images 路径列表中推断并加载 mask_index.json 文件。 Args: scene_id: 场景 ID images: mask 图像路径列表（来自 bbox_json 的 images 字段） Returns:

Positionally encode points that are normalized to [0,1].

Predict masks for the given input prompts, using the currently set image. Input prompts are batched torch tensors and are expected to

inference_states: List of inference states, each state corresponds to a different set of objects.

Reset all tracking inputs and results across the videos.

Run the memory encoder on `high_res_masks`. This is usually after applying non-overlapping constraints to object scores. Since their

Perform inference on a single frame and get its inference results. This would also update `inference_state`.

Add a new object to SAM2 inference states.

Remove an object from SAM2 inference states. This would remove the object from all frames in the video.

Add new mask to a frame.

Batched version of box area. Boxes should be in [x0, y0, x1, y1] format. Inputs: - boxes: Tensor of shape (..., 4) Returns: - a

构造"站在物体A，面向物体B，问物体C的方位和距离"QA（MCQ格式）。 Args: instA, instB, instC: bbox实例字典，需包含 "bounding_box" 字段（8个3D点）。 rng: 随机数生成器，用于生成

Build the SAM3 Tracker module for video tracking. Returns: Sam3TrackerPredictor: Wrapped SAM3 Tracker module

Returns list of utf-8 byte and a corresponding list of unicode strings. The reversible bpe codes work on unicode strings. This means you

分类为8个方向之一（与 mcq_direction_chatgpt.py 一致）。 Returns: (label, idx, yaw_deg) 或 (None, None, None) 如果应该跳过

Remove all input points or mask in a specific frame for a given object.

Remove all input points or mask in all frames throughout the video.

计算 OBB（定向包围盒），返回 (transform, extents)。 Returns: transform: 4x4 变换矩阵（从 OBB 局部坐标系到世界坐标系） extents: (3,) OBB 的尺寸

转换单个QA JSON文件为llamafactory格式

center: (3,) R: (3,3) with columns as local axes in world coords extents_full: (3,) full lengths along each local axis returns: (8,3)

下采样点云到指定的点数 Args: points: 点云坐标 (N, 3) colors: 点云颜色 (N, 3) target_num_points: 目标点数，默认 20 million Returns

在原图上画2D边界框

Args: box_coord (tuple): a tuple containing x0, y0, x1, y1 coordinates, where x0 and y0 are the coordinates of th

在原图上画mask overlay（轮廓）

在原图上画点

Computes the Euclidean Distance Transform (EDT) of a batch of binary images. Args: data: A tensor of shape (B, H, W) representing a

从mask路径中提取图片名称（不含扩展名） 例如: sam_masks_debug/0a7cc12c0e/DSC05865/bed.png -> DSC05865 例如: sam_masks_debug/c285c82ade/02dd3b53_DSC07112/blind_

用中位数+MAD 计算逐轴 Z 分数，删除 |Z| 最大的 10% 点。 这样不依赖固定阈值，按比例裁掉尾部离群。

查找原始图片路径 scannetppv2的目录结构： scannetppv2/ data/ {scene_id}/ dslr/ resized_undistorted_images/

Positionally encode points that are not normalized to [0,1].

为每个QA条目生成唯一的后缀，避免文件名冲突。 基于question_type、sub_question_type、marker_type、ins_id和颜色信息生成hash。

Returns the positional encoding used to encode point prompts, applied to a dense set of points the shape of the image encoding.

Copied from Plenoxels Continuous learning rate decay function. Adapted from JaxNeRF The returned rate is lr_init when step=0 and lr_fina

Returns: ndarray: the visualized image of shape (H, W, 3) (RGB) in uint8 type. The shape is scale

Convert image file to base64 string and get MIME type

Return set of symbol pairs in a word. Word is represented as tuple of symbols (symbols being variable-length strings).

Get relative positional embeddings according to the relative positions of query and key sizes. Args: q_size (int): size of qu

Given a padding mask (following pytorch convention, 1s for padded values), returns whether the padding is on the right or not.

根据标签生成稳定的伪随机颜色。

根据标签生成稳定的伪随机颜色。

保证mask key顺序稳定且可读.

Compute the bounding boxes around the provided masks The masks should be in format [N, H, W] where N is the number of masks, (H, W) are the spati