File size: 7,544 Bytes
53c10a4 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 | from __future__ import annotations
import importlib
import importlib.util
from typing import Any
import numpy as np
import pytest
from orarl.rewards import RewardRouter, RouterSettings, TaskFamily
SEGMENTATION_GT = {
"boxes": [100, 100, 400, 400],
"positive_points": [[150, 150], [250, 250], [350, 350]],
"negative_points": [[0, 0], [700, 700], [900, 100]],
}
TASK_CASES = [
(
{"problem_type": "temporal grounding", "ground_truth": {"time": [1, 3]}},
"<answer>8 to 9</answer>",
),
(
{
"problem_type": "tracking",
"ground_truth": {
"boxes": {
"1": [0, 0, 10, 10],
"2": [10, 10, 20, 20],
}
},
},
'{"boxes":{"1":[0,0,10,10],"2":[10,10,20,20]}}',
),
(
{
"problem_type": "segmentation",
"data_type": "image",
"ground_truth": SEGMENTATION_GT,
},
'<answer>{"boxes":[100,100,400,400]}</answer>',
),
(
{
"problem_type": "spatial grounding",
"ground_truth": {"bbox_2d": [10, 20, 40, 60]},
},
"[10,20,40,60]",
),
(
{
"problem_type": "spatial-temporal grounding",
"ground_truth": {
"time": [1, 2],
"boxes": {
"1": [0, 0, 10, 10],
"2": [10, 10, 20, 20],
},
},
},
'<answer>{"time":[1,2],"boxes":{"1":[50,50,60,60]}}</answer>',
),
(
{"problem_type": "object_counting", "ground_truth": "5"},
"<answer>12</answer>",
),
(
{"problem_type": "video_qa_mc", "ground_truth": "<answer>H</answer>"},
"<answer>A</answer>",
),
]
@pytest.mark.parametrize(("sample", "wrong_response"), TASK_CASES)
def test_default_router_oracles_maximize_all_families(sample, wrong_response):
router = RewardRouter()
oracle_response = router.build_oracle_response(sample)
oracle_score = router.compute_reward(sample, oracle_response)
wrong_score = router.compute_reward(sample, wrong_response)
assert oracle_score["overall"] == pytest.approx(1.0)
assert oracle_score["format"] == 1.0
assert wrong_score["overall"] < oracle_score["overall"]
def test_default_paths_are_packaged_and_importable():
paths = RouterSettings().module_paths
assert set(paths) == set(TaskFamily)
assert all(path.startswith("orarl.") for path in paths.values())
assert all(importlib.util.find_spec(path) is not None for path in paths.values())
def test_default_router_never_requests_an_external_example_module():
imported: list[str] = []
def guarded_import(module_path: str):
assert not module_path.startswith("examples.")
imported.append(module_path)
return importlib.import_module(module_path)
router = RewardRouter(importer=guarded_import)
for sample, _ in TASK_CASES:
response = router.build_oracle_response(sample)
assert router.compute_reward(sample, response)["overall"] == pytest.approx(1.0)
assert len(imported) == len(TaskFamily)
assert all(path.startswith("orarl.") for path in imported)
def test_tracking_missing_frame_contributes_zero():
router = RewardRouter()
sample = {
"problem_type": "tracking",
"ground_truth": {
"boxes": {
"1": [0, 0, 10, 10],
"2": [10, 10, 20, 20],
}
},
}
response = '<answer>{"boxes":{"1":[0,0,10,10]}}</answer>'
score = router.compute_reward(sample, response)
assert score["miou"] == pytest.approx(0.5)
assert score["overall"] == pytest.approx(0.5)
def test_stvg_combines_temporal_iou_with_strict_spatial_iou():
router = RewardRouter()
sample = {
"problem_type": "stvg",
"ground_truth": {
"time": [1, 3],
"boxes": {
"1": [0, 0, 10, 10],
"2": [10, 10, 20, 20],
},
},
}
response = '<answer>{"time":[1,3],"boxes":{"1":[0,0,10,10]}}</answer>'
score = router.compute_reward(sample, response)
assert score["tiou"] == 1.0
assert score["siou_strict"] == pytest.approx(0.5)
assert score["overall"] == pytest.approx(0.2 * 1.0 + 0.8 * 0.5)
def _rle_counts(mask: np.ndarray) -> list[int]:
flattened = mask.T.reshape(-1)
counts: list[int] = []
previous = 0
run = 0
for value in flattened:
current = int(value)
if current == previous:
run += 1
else:
counts.append(run)
run = 1
previous = current
counts.append(run)
return counts
def _compressed_counts(counts: list[int]) -> str:
encoded: list[str] = []
for index, count in enumerate(counts):
value = count - counts[index - 2] if index > 2 else count
more = True
while more:
code = value & 0x1F
value >>= 5
more = value != (-1 if code & 0x10 else 0)
if more:
code |= 0x20
encoded.append(chr(code + 48))
return "".join(encoded)
def _mask_ground_truth() -> dict[str, Any]:
return {
"boxes": [200, 200, 600, 600],
"positive_points": [[300, 300], [400, 400], [500, 500]],
"negative_points": [[50, 50], [800, 800], [950, 500]],
}
def test_segmentation_uses_image_mask_and_proxy_fallback_deterministically():
mask = np.zeros((100, 100), dtype=np.uint8)
mask[20:60, 20:60] = 1
sample = {
"problem_type": "segmentation",
"data_type": "image",
"ground_truth": _mask_ground_truth(),
"segmentation_output": {
"size": [100, 100],
"counts": _rle_counts(mask),
},
}
router = RewardRouter()
response = router.build_oracle_response(sample)
mask_score = router.compute_reward(sample, response)
proxy_sample = {key: value for key, value in sample.items() if key != "segmentation_output"}
first_proxy = router.compute_reward(proxy_sample, response)
second_proxy = router.compute_reward(proxy_sample, response)
assert mask_score["overall"] == pytest.approx(1.0)
assert mask_score["mask_aware_used"] == 1.0
assert first_proxy["overall"] == pytest.approx(1.0)
assert first_proxy == second_proxy
def test_segmentation_decodes_compressed_video_rle_using_media_metadata():
mask = np.zeros((100, 100), dtype=np.uint8)
mask[20:60, 20:60] = 1
empty = np.zeros_like(mask)
target = {**_mask_ground_truth(), "time": 1.0}
sample = {
"problem_type": "segmentation",
"data_type": "video",
"ground_truth": target,
"videos": [{"video": "clip.mp4", "fps": 10}],
"segmentation_output": {
"frames": ["0", "10"],
"segmentation_rle": {
"0": {
"size": [100, 100],
"counts": _compressed_counts(_rle_counts(empty)),
},
"10": {
"size": [100, 100],
"counts": _compressed_counts(_rle_counts(mask)),
},
},
},
}
router = RewardRouter()
response = router.build_oracle_response(sample)
score = router.compute_reward(sample, response)
assert score["overall"] == pytest.approx(1.0)
assert score["mask_aware_used"] == 1.0
assert score["mask_box_iou"] == pytest.approx(1.0)
|