rpx_benchmark.tasks.visual_grounding

Visual grounding: referring expression -> box.

View Source

 1"""Visual grounding: referring expression -> box."""
 2
 3from __future__ import annotations
 4
 5from dataclasses import dataclass
 6
 7from ..api import TaskType
 8from ._pipeline import PipelineResult, TaskRunConfig, run_pipeline
 9from .registry import TaskSpec, register_task
10
11PRIMARY_METRIC = "grounding_acc"
12
13
14@dataclass
15class VisualGroundingRunConfig(TaskRunConfig):
16    pass
17
18
19def run_visual_grounding(cfg: VisualGroundingRunConfig) -> PipelineResult:
20    return run_pipeline(
21        task=TaskType.VISUAL_GROUNDING,
22        primary_metric=PRIMARY_METRIC,
23        cfg=cfg,
24    )
25
26
27TASK_SPEC = TaskSpec(
28    task=TaskType.VISUAL_GROUNDING,
29    display_name="Visual Grounding",
30    description="Predict the box for a referring expression on an RGB frame.",
31    primary_metric=PRIMARY_METRIC,
32    required_modalities=["rgb", "questionnaires"],
33    higher_is_better=True,
34    run=run_visual_grounding,
35)
36
37register_task(TASK_SPEC)

PRIMARY_METRIC = 'grounding_acc'

@dataclass

class VisualGroundingRunConfig(rpx_benchmark.tasks._pipeline.TaskRunConfig): View Source

15@dataclass
16class VisualGroundingRunConfig(TaskRunConfig):
17    pass

VisualGroundingRunConfig(model: 'Optional[BenchmarkableModel]' = None, split: 'Difficulty | str' = , repo_id: 'Optional[str]' = None, cache_dir: 'Optional[str]' = None, revision: 'Optional[str]' = '51432e5d48895bbafb490ae7c5e34e0f6b7da7d6', batch_size: 'int' = 1, require_cuda: 'bool' = False, skip_flops: 'bool' = False, max_samples: 'Optional[int]' = None, manifest_path: 'Optional[str]' = None, device: 'str' = 'cuda', output_dir: 'Optional[str]' = None, progress: 'Optional[ProgressCallback]' = None, metric_suite: 'Optional[MetricSuite]' = None, upload_to_box: 'bool' = False, box_folder_id: 'Optional[str]' = None)

VisualGroundingRunConfig( model: Optional[rpx_benchmark.adapters.BenchmarkableModel] = None, split: rpx_benchmark.api.Difficulty | str = <Difficulty.HARD: 'hard'>, repo_id: Optional[str] = None, cache_dir: Optional[str] = None, revision: Optional[str] = '51432e5d48895bbafb490ae7c5e34e0f6b7da7d6', batch_size: int = 1, require_cuda: bool = False, skip_flops: bool = False, max_samples: Optional[int] = None, manifest_path: Optional[str] = None, device: str = 'cuda', output_dir: Optional[str] = None, progress: Optional[Callable[[int, Optional[int], str], NoneType]] = None, metric_suite: Optional[rpx_benchmark.metrics.MetricSuite] = None, upload_to_box: bool = False, box_folder_id: Optional[str] = None)

Inherited Members

rpx_benchmark.tasks._pipeline.TaskRunConfig: model; split; repo_id; cache_dir; revision; batch_size; require_cuda; skip_flops; max_samples; manifest_path; device; output_dir; progress; metric_suite; upload_to_box; box_folder_id

def run_visual_grounding( cfg: VisualGroundingRunConfig) -> Tuple[rpx_benchmark.metrics.registry.BenchmarkResult, rpx_benchmark.deployment.DeploymentReadinessReport, Dict[str, pathlib.Path]]: View Source

20def run_visual_grounding(cfg: VisualGroundingRunConfig) -> PipelineResult:
21    return run_pipeline(
22        task=TaskType.VISUAL_GROUNDING,
23        primary_metric=PRIMARY_METRIC,
24        cfg=cfg,
25    )

TASK_SPEC = TaskSpec(task=<TaskType.VISUAL_GROUNDING: 'visual_grounding'>, display_name='Visual Grounding', primary_metric='grounding_acc', required_modalities=['rgb', 'questionnaires'], run=<function run_visual_grounding>, higher_is_better=True, description='Predict the box for a referring expression on an RGB frame.', temporal_stability_fn=None, geometric_coherence_fn=None)