Structure to hold parameters for the image processor. This structure contains various settings for configuring the image processing. More...
#include <CommonParameters.h>
Public Attributes | |
| std::filesystem::path | inference_plugin |
| int | n_tiles_x {1} |
| Number of tiles in horizontal direction. More... | |
| int | n_tiles_y {1} |
| Number of tiles in vertical direction. More... | |
| float | ol_x {0.0f} |
| Overlap of tiles in horizontal direction as a fraction of tile width. More... | |
| float | ol_y {0.0f} |
| Overlap of tiles in vertical direction as a fraction of tile height. More... | |
| bool | swapRB {false} |
| Swap Red and Blue channels. Useful when input data is in BGR format (when its loaded with OpenCV for example), because it is done fast in the processor. More... | |
| cv::Rect_< float > | roi {0.0f, 0.0f, 1.0f, 1.0f} |
| Region of interest. Measured from 0 to 1 The region of interest is a rectangle that defines the area of the image that will be processed. In format of x, y, width, height. More... | |
| ModelThresholds | thresholds |
| The thresholds for the model. More... | |
| float | nms_threshold {0.5f} |
| The NMS threshold (Non-Maximum Suppression threshold) is used during the process to eliminate redundant detections. NMS is a technique used to select the best bounding box when multiple boxes overlap and detect the same object. The NMS threshold is a value between 0 and 1 that represents the Intersection over Union (IoU) threshold. IoU is a measure of the overlap between two bounding boxes. More... | |
| int | queue_size {std::numeric_limits<int>::max()} |
| Specifies the maximum size of the queue. More... | |
| PerTypeProcessingConfig | per_type_processing_config {DEFAULT_PER_TYPE_PROCESSING_CONFIG} |
| Fine-tuning the processing of detections based on their class. More... | |
Detailed Description
Structure to hold parameters for the image processor. This structure contains various settings for configuring the image processing.
Member Data Documentation
◆ inference_plugin
| std::filesystem::path celantur::ProcessorParams::inference_plugin |
Path to the inference plugin For CPU model it is libONNXInference.so or libOpenVINOInference.so For GPU model it is libTensorRTRuntime.so
◆ n_tiles_x
| int celantur::ProcessorParams::n_tiles_x {1} |
Number of tiles in horizontal direction.
◆ n_tiles_y
| int celantur::ProcessorParams::n_tiles_y {1} |
Number of tiles in vertical direction.
◆ nms_threshold
| float celantur::ProcessorParams::nms_threshold {0.5f} |
The NMS threshold (Non-Maximum Suppression threshold) is used during the process to eliminate redundant detections. NMS is a technique used to select the best bounding box when multiple boxes overlap and detect the same object. The NMS threshold is a value between 0 and 1 that represents the Intersection over Union (IoU) threshold. IoU is a measure of the overlap between two bounding boxes.
- Note
- Probably not important for the end user, but in some cases might become useful.
◆ ol_x
| float celantur::ProcessorParams::ol_x {0.0f} |
Overlap of tiles in horizontal direction as a fraction of tile width.
◆ ol_y
| float celantur::ProcessorParams::ol_y {0.0f} |
Overlap of tiles in vertical direction as a fraction of tile height.
◆ per_type_processing_config
| PerTypeProcessingConfig celantur::ProcessorParams::per_type_processing_config {DEFAULT_PER_TYPE_PROCESSING_CONFIG} |
Fine-tuning the processing of detections based on their class.
◆ queue_size
| int celantur::ProcessorParams::queue_size {std::numeric_limits<int>::max()} |
Specifies the maximum size of the queue.
This parameter defines the maximum number of elements that can be held in the processing queue of CelanturSDK::Processor at any given time. By default, it is set to the
largest possible integer value, effectively making the queue unbound unless explicitly configured.
- Note
- Adjusting the queue size can help manage memory usage and processing latency, especially in high-throughput scenarios.
◆ roi
| cv::Rect_<float> celantur::ProcessorParams::roi {0.0f, 0.0f, 1.0f, 1.0f} |
Region of interest. Measured from 0 to 1 The region of interest is a rectangle that defines the area of the image that will be processed. In format of x, y, width, height.
◆ swapRB
| bool celantur::ProcessorParams::swapRB {false} |
Swap Red and Blue channels. Useful when input data is in BGR format (when its loaded with OpenCV for example), because it is done fast in the processor.
◆ thresholds
| ModelThresholds celantur::ProcessorParams::thresholds |
The thresholds for the model.
