mmcv.ops¶
Border align pooling layer. 

CARAFE: ContentAware ReAssembly of FEatures 

A unified package of CARAFE upsampler that contains: 1) channel compressor 2) content encoder 3) CARAFE op. 

alias of 

alias of 

Corner Pooling. 

Correlation operator 

CrissCross Attention Module. 

Deformable 2D convolution. 

A Deformable Conv Encapsulation that acts as normal Conv layers. 

Scatters points into voxels, used in the voxel encoder with dynamic voxelization. 

Fused bias leaky ReLU. 

Group xyz with feature. 

alias of 

A MaskedConv2d which inherits the official Conv2d. 

alias of 

A ModulatedDeformable Conv Encapsulation that acts as normal Conv layers. 

An attention module used in DeformableDetr. 

Points sampling. 

The operation of precision RoI pooling. 

Groups points with a ball query of radius. 

Rotationinvariant RoI align pooling layer for rotated proposals. 

RoI align pooling layer. 

RoI align pooling layer for rotated proposals. 

Encode the geometryspecific features of each 3D proposal. 

Encode the geometryspecific features of each 3D proposal. 

SAC (Switchable Atrous Convolution) 

place holder, All module subclass from this will take sptensor in SparseSequential. 

A sequential container. 

Synchronized Batch Normalization. 

Temporal Interlace Shift. 

Convert kitti points(N, >=3) to voxels. 
Performs nonmaximum suppression in a batched fashion. 

Calculate overlap between two set of bboxes. 

Return intersectionoverunion (Jaccard index) of boxes. 

Calculate boxes 3D IoU. 

Calculate boxes IoU in the Bird’s Eye View. 

Calculate boxes BEV overlap. 

Expand kernel contours so that foreground pixels are assigned into instances. 

Return generalized intersectionoverunion (Jaccard index) between point sets and polygons. 

Return intersectionoverunion (Jaccard index) between point sets and polygons. 

Calculate differentiable iou of rotated 2d boxes. 

Calculate differentiable iou of rotated 3d boxes. 

Fused bias leaky ReLU function. 

Find the smallest polygons that surrounds all points in the point sets. 

Dispatch to either CPU or GPU NMS implementations. 

3D NMS function GPU implementation (for BEV boxes). 

Normal 3D NMS function GPU implementation. 

NMS function GPU implementation (for BEV boxes). 

Matched dets into different groups by NMS. 

Normal NMS function GPU implementation (for BEV boxes). 

Performs nonmaximum suppression (NMS) on the rotated boxes according to their intersectionoverunion (IoU). 

Group pixels into text instances, which is widely used text detection methods. 

A wrapper around 

Find all boxes in which each point is (CUDA). 

Find all boxes in which each point is (CPU). 

Find the box in which each point is (CUDA). 

Judging whether points are inside polygons, which is used in the ATSS assignment for the rotated boxes. 

Convert roi based relative point coordinates to image based absolute point coordinates. 

pytorch edition of tensorflow scatter_nd. 

Dispatch to only CPU Soft NMS implementations. 

Pad, upsample, filter, and downsample a batch of 2D images. 
