rpx_benchmark.tasks.detection

Object detection (closed vocab + open vocab).

 1"""Object detection (closed vocab + open vocab)."""
 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 = "f1"
12
13
14@dataclass
15class ObjectDetectionRunConfig(TaskRunConfig):
16    pass
17
18
19def run_object_detection(cfg: ObjectDetectionRunConfig) -> PipelineResult:
20    return run_pipeline(
21        task=TaskType.OBJECT_DETECTION,
22        primary_metric=PRIMARY_METRIC,
23        cfg=cfg,
24    )
25
26
27def run_open_vocab_detection(cfg: ObjectDetectionRunConfig) -> PipelineResult:
28    return run_pipeline(
29        task=TaskType.OPEN_VOCAB_DETECTION,
30        primary_metric=PRIMARY_METRIC,
31        cfg=cfg,
32    )
33
34
35TASK_SPEC = TaskSpec(
36    task=TaskType.OBJECT_DETECTION,
37    display_name="Object Detection",
38    description="Detect and label boxes on a single RGB frame (closed vocab).",
39    primary_metric=PRIMARY_METRIC,
40    required_modalities=["rgb", "boxes"],
41    higher_is_better=True,
42    run=run_object_detection,
43)
44
45OPEN_VOCAB_TASK_SPEC = TaskSpec(
46    task=TaskType.OPEN_VOCAB_DETECTION,
47    display_name="Open-Vocabulary Detection",
48    description="Detect objects given a free-text vocabulary.",
49    primary_metric=PRIMARY_METRIC,
50    required_modalities=["rgb", "boxes"],
51    higher_is_better=True,
52    run=run_open_vocab_detection,
53)
54
55register_task(TASK_SPEC)
56register_task(OPEN_VOCAB_TASK_SPEC)
PRIMARY_METRIC = 'f1'
@dataclass
class ObjectDetectionRunConfig(rpx_benchmark.tasks._pipeline.TaskRunConfig): 
15@dataclass
16class ObjectDetectionRunConfig(TaskRunConfig):
17    pass

ObjectDetectionRunConfig(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)

ObjectDetectionRunConfig( 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_object_detection( cfg: ObjectDetectionRunConfig) -> Tuple[rpx_benchmark.metrics.registry.BenchmarkResult, rpx_benchmark.deployment.DeploymentReadinessReport, Dict[str, pathlib.Path]]: 
20def run_object_detection(cfg: ObjectDetectionRunConfig) -> PipelineResult:
21    return run_pipeline(
22        task=TaskType.OBJECT_DETECTION,
23        primary_metric=PRIMARY_METRIC,
24        cfg=cfg,
25    )
def run_open_vocab_detection( cfg: ObjectDetectionRunConfig) -> Tuple[rpx_benchmark.metrics.registry.BenchmarkResult, rpx_benchmark.deployment.DeploymentReadinessReport, Dict[str, pathlib.Path]]: 
28def run_open_vocab_detection(cfg: ObjectDetectionRunConfig) -> PipelineResult:
29    return run_pipeline(
30        task=TaskType.OPEN_VOCAB_DETECTION,
31        primary_metric=PRIMARY_METRIC,
32        cfg=cfg,
33    )
TASK_SPEC = TaskSpec(task=<TaskType.OBJECT_DETECTION: 'object_detection'>, display_name='Object Detection', primary_metric='f1', required_modalities=['rgb', 'boxes'], run=<function run_object_detection>, higher_is_better=True, description='Detect and label boxes on a single RGB frame (closed vocab).', temporal_stability_fn=None, geometric_coherence_fn=None)
OPEN_VOCAB_TASK_SPEC = TaskSpec(task=<TaskType.OPEN_VOCAB_DETECTION: 'open_vocab_detection'>, display_name='Open-Vocabulary Detection', primary_metric='f1', required_modalities=['rgb', 'boxes'], run=<function run_open_vocab_detection>, higher_is_better=True, description='Detect objects given a free-text vocabulary.', temporal_stability_fn=None, geometric_coherence_fn=None)