|
ReUseX
0.0.5
3D Point Cloud Processing for Building Reuse
|
#include <IDataset.hpp>


Public Types | |
| using | Pair = std::pair<std::unique_ptr<IData>, size_t> |
Public Member Functions | |
| IDataset (std::shared_ptr< ProjectDB > database) | |
| IDataset (std::filesystem::path dbPath) | |
| virtual | ~IDataset ()=default |
| size_t | size () const |
| int | node_id (const std::size_t index) const |
| size_t | filter_annotated () |
| Remove already-annotated frames from the dataset. | |
| size_t | filter_annotated_prefix () |
| Remove only the leading contiguous run of already-annotated frames. | |
| virtual Pair | get (const std::size_t index) const =0 |
| virtual bool | save (const std::span< Pair > &data)=0 |
| void | set_confidence (float confidence) |
| void | set_prompts (std::vector< std::string > prompts) |
Protected Member Functions | |
| cv::Mat | image (const std::size_t index) const |
| bool | save_image (const std::size_t index, const cv::Mat &image) |
| std::shared_ptr< ProjectDB > | database () const |
Protected Attributes | |
| float | confidence_ = 0.5f |
| Detection confidence threshold for get(). | |
| std::vector< std::string > | prompts_ |
| Concept prompts for get() (empty=default). | |
Definition at line 36 of file IDataset.hpp.
| using reusex::vision::IDataset::Pair = std::pair<std::unique_ptr<IData>, size_t> |
Definition at line 44 of file IDataset.hpp.
|
explicit |
References database().
|
explicit |
|
virtualdefault |
|
protected |
Referenced by IDataset().
| size_t reusex::vision::IDataset::filter_annotated | ( | ) |
Remove already-annotated frames from the dataset.
Queries the database for frames that already have segmentation images and removes them from this dataset's ID list so they won't be processed.
| size_t reusex::vision::IDataset::filter_annotated_prefix | ( | ) |
Remove only the leading contiguous run of already-annotated frames.
Unlike filter_annotated(), which drops every already-annotated frame regardless of position, this removes only the maximal prefix of the ordered id list whose frames all already have segmentation images, stopping at the first unannotated frame. This is the correct resume behaviour for the stateful video-tracker path: dropping mid-sequence frames would leave gaps in the temporal memory bank, so we skip only the already-completed prefix and re-process everything from the first gap onward (rebuilding memory from that boundary). For the stateless default path it degrades gracefully to the same "skip completed leading frames" behaviour.
|
pure virtual |
|
protected |
Referenced by save_image().
| int reusex::vision::IDataset::node_id | ( | const std::size_t | index | ) | const |
|
pure virtual |
|
protected |
References image().
|
inline |
Definition at line 152 of file IDataset.hpp.
References confidence_.
|
inline |
Definition at line 158 of file IDataset.hpp.
References prompts_.
| size_t reusex::vision::IDataset::size | ( | ) | const |
|
protected |
Detection confidence threshold for get().
Definition at line 163 of file IDataset.hpp.
Referenced by set_confidence().
|
protected |
Concept prompts for get() (empty=default).
Definition at line 165 of file IDataset.hpp.
Referenced by set_prompts().