Called upon thread destruction (not main thread!)

* @brief Returns the number of channels in the image tensor. * * @return The number of channels. */

* @brief Returns the number of planes in the planar image tensor. * * @return The number of planes. */

* @brief Returns the number of rows in the image tensor. * * @return The number of rows. */

* @brief Returns the number of samples in the tensor. * * @return The number of samples. */

* @brief Retrieves the number of spatial dimensions in the tensor layout. * @return Number of spatial dimensions. */

Transfer NumPy to GPU using CUDA Python, then create CV-CUDA tensor. For more details on CUDA Python interop, see cuda_python_interop_1.py

Transfer NumPy to GPU using CuPy, then create CV-CUDA tensor. For more details on CuPy interop, see cupy_interop.py

Transfer NumPy to GPU using PyCUDA, then create CV-CUDA tensor. For more details on PyCUDA interop, see pycuda_interop.py

Transfer NumPy to GPU using PyTorch, then create CV-CUDA tensor. For more details on PyTorch interop, see pytorch_interop.py

Parses the nvtx_gpu_proj_trace JSON generated by NSYS and returns a dictionary keyed by process_id, thread_id and range_id. The values are va

Parses the nvtx_pushpop_trace JSON generated by NSYS and returns a dictionary keyed by process_id, thread_id and range_id. The values are var

Parses and validates the values of the default command line arguments.

* @brief Computes the stride for the plane dimension. * * @return The plane stride, or 0 if the plane dimension is not present or is not fir

Plots GPU metrics as a matplotlib plot.

Relinquishes the ownership of the handle and returns the formerly managed handle. * * The function returns the handle and stops managing it.

Replaces the currently owned descriptor with the one passed in the argument. * * The ownership of the descriptor is transfered to `CustomMem

Resets the GPU clocks.

* @brief Retrieves a pointer to the data of a specific row. * * @param y The row index. * @return A pointer to the row's data. */

* @brief Computes the stride for the row (or height) dimension. * * @return The row stride, or 0 if the height dimension is not present.

Runs the operator and returns the result. :param input: The input tensor to run the operator on. :returns: The result from th

Runs the per operator benchmarks. It automatically discovers eligible operators for benchmarking, sets them up, runs them and saves the runti

Performs various setup activities to set this operator before it can be run. :param input: The input tensor to run the operator on.

returns number of contours

note: this routine clips fills that extend off the edges... ideally this wouldn't happen, but it could happen if the truetype glyph bounding boxes are