合并各场景临时jsonl，重排全局id.

Args: - pred_probs: (num_det,) float Tensor, containing the score (probability) of each detection - pred_masks: (num_det, H_mask, W_m

Perform NMS given the iou matrix, the scores and the iou threshold Args: ious: Pairwise IoU tensor of shape (N, N). scores: Score

Returns: (H, W) array, a mask for all pixels that have a prediction

Restrict angles to the range of [-180, 180) degrees

计算两个 OBB 的交叠体积，判断是否重叠。 Args: transform_a: 4x4 变换矩阵 extents_a: (3,) OBB 尺寸 transform_b: 4x4 变换矩阵 extents_b: (3

计算两个 OBB 的交叠体积，判断是否重叠。 Args: transform_a: 4x4 变换矩阵 extents_a: (3,) OBB 尺寸 transform_b: 4x4 变换矩阵 extents_b: (3

Args: boxes (ndarray): an Nx5 numpy array of (x_center, y_center, width, height, angle_degrees) format

Return intersection-over-union (Jaccard index) of boxes. Both sets of boxes are expected to be in (x_center, y_center, width, height, an

Parse the actions in history before the last propagation and prepare for the next propagation. We support multiple actions (add/remov

Post-process a mesh to filter out floaters and disconnected parts

Post-process a mesh to filter out floaters and disconnected parts

Perform PostProcessing on output masks.

Predict masks for the given input prompts, using the currently set image. Arguments: point_coords (np.ndarray or None): A

This function is very similar to predict(...), however it is used for batched mode, when the model is expected to generate predictions on multiple ima

Predicts masks. See 'forward' for more details.

This method is only used for benchmark eval (not used in the demo).

Rasterize the polygons into a mask image and crop the mask content in the given box. The cropped mask is resized to (mask_size, mask_size

Read cameras from DL3DV format DL3DV structure: - dense/cam/ directory: contains .npz files with intrinsics and pose - dense/rgb/ or

Read cameras from ScanNet format ScanNet structure: - intrinsic/intrinsic_color.txt: shared color intrinsics - pose/ directory: conta

Read cameras from ScanNetV2 format. ScanNetV2 structure: - color/ directory: image files - pose/ directory: c2w txt files (4x4) -

see: src/base/reconstruction.cc void Reconstruction::ReadImagesBinary(const std::string& path) void Reconstruction::WriteImagesBi

Taken from https://github.com/colmap/colmap/blob/dev/scripts/python/read_write_model.py

see: src/base/reconstruction.cc void Reconstruction::WriteCamerasBinary(const std::string& path) void Reconstruction::ReadCameras

Taken from https://github.com/colmap/colmap/blob/dev/scripts/python/read_write_model.py

see: src/base/reconstruction.cc void Reconstruction::ReadPoints3DBinary(const std::string& path) void Reconstruction::WritePoints

see: src/base/reconstruction.cc void Reconstruction::ReadPoints3DText(const std::string& path) void Reconstruction::WritePoints3D

利用中值滤波删除深度飞点，返回与输入同尺寸的米单位深度。 飞点会被置为 0。

利用中值滤波删除深度飞点，返回与输入同尺寸的米单位深度。 飞点会被置为 0。

We try to remove object from tracker states on every GPU, it will do nothing for states without this object.

Greedy keep: sort by score desc; keep a mask if IoM to all kept masks <= threshold. If pred_masks has length 0 or 1, returns sample unchanged

使用 3DGS 渲染深度与颜色，不再初始化完整的 Scene，而是直接构造 Camera。 注意：对于 DL3DV 场景，应该使用 load_da3_depths 而不是这个函数。

使用 3DGS 渲染深度与颜色，不再初始化完整的 Scene，而是直接构造 Camera。

Reset the session to its initial state (as when it's initial opened).

QA 里 image_a/image_b 一般是 "DSCxxxxx.JPG" 或 "0.jpg" 等。 支持的路径格式: - scannetppv2: <data_root>/<scene_id>/dslr/resized_undistorted_images/<im

Makes a function retry itself after encountering pytorch's CUDA OOM error. It will first retry after calling `torch.cuda.empty_cache()`.

Run SAM 3 image inference with text prompts and format the outputs

采样在图像对的两个图像中同时出现的对象。返回 [(image_pair, instance), ...]

采样在图像对的 image A 中出现的对象。返回 [(image_pair, instance), ...]

从图像对中采样 (object A, object B, region C) 三元组。 约束：region C 必须来自 image B。 约束：region C 的 bbox 不能包含 object A 的中心点。

从图像对中采样 (object A, object B, region C) 三元组。 约束：region C 必须来自 image B。 用于 proxy QA，约束：region C 的 bbox 不能包含 object B 的中心点。