ReUseX  0.0.5
3D Point Cloud Processing for Building Reuse
Loading...
Searching...
No Matches
reusex::vision::DescribeConfig Struct Reference

Options for the VLM attribute-extraction stage. More...

#include <describe.hpp>

Collaboration diagram for reusex::vision::DescribeConfig:

Public Attributes

std::string api_url = "http://localhost:11434/v1"
 OpenAI-compatible chat/completions base URL.
std::string model = "qwen2.5vl"
 Model name the endpoint routes to.
std::string api_key
 Optional bearer token (cloud providers). Empty = no Authorization header.
std::string instances_cloud = "instances"
 Name of the instance-label cloud used to resolve a material's points.
std::string point_cloud = "cloud"
 Name of the XYZRGB cloud index-aligned with instances_cloud.
std::string prompt
 Instruction sent alongside each crop.
bool skip_existing = false
 Skip materials that already have a stored annotation (resume runs).
int crop_padding = 8
 Pixels of padding around the projected instance bbox before cropping.
int min_view_points = 20
 Minimum projected in-bounds points for a frame to be a usable view.
std::size_t max_points_per_frame = 5000
 Max points to reproject per frame (subsample large instances for speed).
int connect_timeout_s = 10
 CURL TCP connect timeout in seconds.
int total_timeout_s = 120
 CURL total request timeout in seconds (covers connect + send + receive).

Detailed Description

Options for the VLM attribute-extraction stage.

CLI flags mirror these defaults (STANDARDS ยง4).

Definition at line 23 of file describe.hpp.

Member Data Documentation

◆ api_key

std::string reusex::vision::DescribeConfig::api_key

Optional bearer token (cloud providers). Empty = no Authorization header.

Definition at line 30 of file describe.hpp.

◆ api_url

std::string reusex::vision::DescribeConfig::api_url = "http://localhost:11434/v1"

OpenAI-compatible chat/completions base URL.

Default is a local Ollama. The /chat/completions path is appended if not already present.

Definition at line 26 of file describe.hpp.

◆ connect_timeout_s

int reusex::vision::DescribeConfig::connect_timeout_s = 10

CURL TCP connect timeout in seconds.

0 = no limit. VLM servers can be slow to accept; 10 s catches a wedged endpoint fast.

Definition at line 49 of file describe.hpp.

◆ crop_padding

int reusex::vision::DescribeConfig::crop_padding = 8

Pixels of padding around the projected instance bbox before cropping.

Definition at line 42 of file describe.hpp.

◆ instances_cloud

std::string reusex::vision::DescribeConfig::instances_cloud = "instances"

Name of the instance-label cloud used to resolve a material's points.

Materials link to instances in THIS cloud (via instance_materials).

Definition at line 33 of file describe.hpp.

◆ max_points_per_frame

std::size_t reusex::vision::DescribeConfig::max_points_per_frame = 5000

Max points to reproject per frame (subsample large instances for speed).

Definition at line 46 of file describe.hpp.

◆ min_view_points

int reusex::vision::DescribeConfig::min_view_points = 20

Minimum projected in-bounds points for a frame to be a usable view.

Definition at line 44 of file describe.hpp.

◆ model

std::string reusex::vision::DescribeConfig::model = "qwen2.5vl"

Model name the endpoint routes to.

Definition at line 28 of file describe.hpp.

◆ point_cloud

std::string reusex::vision::DescribeConfig::point_cloud = "cloud"

Name of the XYZRGB cloud index-aligned with instances_cloud.

Definition at line 35 of file describe.hpp.

◆ prompt

std::string reusex::vision::DescribeConfig::prompt

Instruction sent alongside each crop.

A default prompt requesting a description + attributes object is used when empty.

Definition at line 38 of file describe.hpp.

◆ skip_existing

bool reusex::vision::DescribeConfig::skip_existing = false

Skip materials that already have a stored annotation (resume runs).

Definition at line 40 of file describe.hpp.

◆ total_timeout_s

int reusex::vision::DescribeConfig::total_timeout_s = 120

CURL total request timeout in seconds (covers connect + send + receive).

0 = no limit. 120 s is generous enough for large-image VLM responses.

Definition at line 52 of file describe.hpp.


The documentation for this struct was generated from the following file: