Skip to content

Commit 2071af7

Browse files
leeclemnetleeclemnet
committed
update tests
1 parent 8ec678b commit 2071af7

2 files changed

Lines changed: 148 additions & 65 deletions

File tree

‎inference_models/tests/integration_tests/models/conftest.py‎

Lines changed: 22 additions & 4 deletions
Original file line numberDiff line numberDiff line change
@@ -73,7 +73,9 @@
7373
COIN_COUNTING_YOLACT_ONNX_STATIC_BS_STATIC_CROP_STRETCH_URL = "https://storage.googleapis.com/roboflow-tests-assets/rf-platform-models/yolact-static-bs-static-crop-stretch-onnx.zip"
7474
COIN_COUNTING_YOLACT_ONNX_STATIC_BS_STRETCH_URL = "https://storage.googleapis.com/roboflow-tests-assets/rf-platform-models/yolact-static-bs-stretch-onnx.zip"
7575

76-
COIN_COUNTING_YOLOLITE_N_ONNX_DYNAMIC_BS_LETTERBOX_URL = "https://storage.googleapis.com/roboflow-tests-assets/rf-platform-models/coin-counting-yololite-n-onnx-dynamic-bs-letterbox.zip"
76+
COIN_COUNTING_YOLOLITE_EDGE_N_ONNX_STATIC_BS_STRETCH_URL = "https://storage.googleapis.com/roboflow-tests-assets/rf-platform-models/coin-counting-yololite-edge-n-onnx-static-bs-stretch.zip"
77+
COIN_COUNTING_YOLOLITE_EDGE_N_ONNX_DYNAMIC_BS_STRETCH_URL = "https://storage.googleapis.com/roboflow-tests-assets/rf-platform-models/coin-counting-yololite-edge-n-onnx-dynamic-bs-stretch.zip"
78+
COIN_COUNTING_YOLOLITE_EDGE_N_ONNX_DYNAMIC_BS_STRETCH_FUSED_NMS_URL = "https://storage.googleapis.com/roboflow-tests-assets/rf-platform-models/coin-counting-yololite-edge-n-onnx-dynamic-bs-stretch-fused-nms.zip"
7779

7880
ASL_YOLOV8N_SEG_ONNX_DYNAMIC_BS_STRETCH_URL = "https://storage.googleapis.com/roboflow-tests-assets/rf-platform-models/yolov8n-seg-onnx-dynamic-bs-stretch.zip"
7981
ASL_YOLOV8N_SEG_ONNX_DYNAMIC_BS_STRETCH_FUSED_NMS_URL = "https://storage.googleapis.com/roboflow-tests-assets/rf-platform-models/yolov8n-seg-onnx-dynamic-bs-stretch-fused-nms.zip"
@@ -700,10 +702,26 @@ def coin_counting_yolo_nas_onnx_static_bs_center_crop_package() -> str:
700702

701703

702704
@pytest.fixture(scope="module")
703-
def coin_counting_yololite_n_onnx_dynamic_bs_letterbox_package() -> str:
705+
def coin_counting_yololite_edge_n_onnx_static_bs_stretch_package() -> str:
704706
return download_model_package(
705-
model_package_zip_url=COIN_COUNTING_YOLOLITE_N_ONNX_DYNAMIC_BS_LETTERBOX_URL,
706-
package_name="coin-counting-yololite-n-onnx-dynamic-bs-letterbox",
707+
model_package_zip_url=COIN_COUNTING_YOLOLITE_EDGE_N_ONNX_STATIC_BS_STRETCH_URL,
708+
package_name="coin-counting-yololite-edge-n-onnx-static-bs-stretch",
709+
)
710+
711+
712+
@pytest.fixture(scope="module")
713+
def coin_counting_yololite_edge_n_onnx_dynamic_bs_stretch_package() -> str:
714+
return download_model_package(
715+
model_package_zip_url=COIN_COUNTING_YOLOLITE_EDGE_N_ONNX_DYNAMIC_BS_STRETCH_URL,
716+
package_name="coin-counting-yololite-edge-n-onnx-dynamic-bs-stretch",
717+
)
718+
719+
720+
@pytest.fixture(scope="module")
721+
def coin_counting_yololite_edge_n_onnx_dynamic_bs_stretch_fused_nms_package() -> str:
722+
return download_model_package(
723+
model_package_zip_url=COIN_COUNTING_YOLOLITE_EDGE_N_ONNX_DYNAMIC_BS_STRETCH_FUSED_NMS_URL,
724+
package_name="coin-counting-yololite-edge-n-onnx-dynamic-bs-stretch-fused-nms",
707725
)
708726

709727

‎inference_models/tests/integration_tests/models/test_yololite_object_detection_predictions_onnx.py‎

Lines changed: 126 additions & 61 deletions
Original file line numberDiff line numberDiff line change
@@ -3,32 +3,58 @@
33
import torch
44

55

6+
# ── Static non-fused (batch=1) ──────────────────────────────────────────────
7+
8+
69
@pytest.mark.slow
710
@pytest.mark.onnx_extras
8-
def test_onnx_package_with_dynamic_batch_size_and_letterbox_numpy(
9-
coin_counting_yololite_n_onnx_dynamic_bs_letterbox_package: str,
11+
def test_static_non_fused_numpy(
12+
coin_counting_yololite_edge_n_onnx_static_bs_stretch_package: str,
1013
coins_counting_image_numpy: np.ndarray,
1114
) -> None:
12-
# given
1315
from inference_models.models.yololite.yololite_object_detection_onnx import (
1416
YOLOLiteForObjectDetectionOnnx,
1517
)
1618

1719
model = YOLOLiteForObjectDetectionOnnx.from_pretrained(
18-
model_name_or_path=coin_counting_yololite_n_onnx_dynamic_bs_letterbox_package,
20+
model_name_or_path=coin_counting_yololite_edge_n_onnx_static_bs_stretch_package,
1921
onnx_execution_providers=["CUDAExecutionProvider", "CPUExecutionProvider"],
2022
)
23+
predictions = model(
24+
coins_counting_image_numpy, confidence=0.25, iou_threshold=0.45
25+
)
26+
27+
assert len(predictions) == 1
28+
assert predictions[0].xyxy.shape[1] == 4
29+
assert len(predictions[0].confidence) > 0
30+
assert torch.all(predictions[0].confidence >= 0.25)
31+
assert torch.all(predictions[0].confidence <= 1.0)
32+
33+
34+
# ── Dynamic non-fused ────────────────────────────────────────────────────────
35+
36+
37+
@pytest.mark.slow
38+
@pytest.mark.onnx_extras
39+
def test_dynamic_non_fused_numpy(
40+
coin_counting_yololite_edge_n_onnx_dynamic_bs_stretch_package: str,
41+
coins_counting_image_numpy: np.ndarray,
42+
) -> None:
43+
from inference_models.models.yololite.yololite_object_detection_onnx import (
44+
YOLOLiteForObjectDetectionOnnx,
45+
)
2146

22-
# when
47+
model = YOLOLiteForObjectDetectionOnnx.from_pretrained(
48+
model_name_or_path=coin_counting_yololite_edge_n_onnx_dynamic_bs_stretch_package,
49+
onnx_execution_providers=["CUDAExecutionProvider", "CPUExecutionProvider"],
50+
)
2351
predictions = model(
2452
coins_counting_image_numpy,
2553
confidence=0.25,
2654
iou_threshold=0.45,
2755
max_detections=100,
2856
)
2957

30-
# then
31-
assert isinstance(predictions, list)
3258
assert len(predictions) == 1
3359
assert predictions[0].xyxy.shape[1] == 4
3460
assert predictions[0].xyxy.dtype == torch.int32
@@ -41,38 +67,30 @@ def test_onnx_package_with_dynamic_batch_size_and_letterbox_numpy(
4167

4268
@pytest.mark.slow
4369
@pytest.mark.onnx_extras
44-
def test_onnx_package_with_dynamic_batch_size_and_letterbox_batch_numpy(
45-
coin_counting_yololite_n_onnx_dynamic_bs_letterbox_package: str,
70+
def test_dynamic_non_fused_batch_numpy(
71+
coin_counting_yololite_edge_n_onnx_dynamic_bs_stretch_package: str,
4672
coins_counting_image_numpy: np.ndarray,
4773
) -> None:
48-
# given
4974
from inference_models.models.yololite.yololite_object_detection_onnx import (
5075
YOLOLiteForObjectDetectionOnnx,
5176
)
5277

5378
model = YOLOLiteForObjectDetectionOnnx.from_pretrained(
54-
model_name_or_path=coin_counting_yololite_n_onnx_dynamic_bs_letterbox_package,
79+
model_name_or_path=coin_counting_yololite_edge_n_onnx_dynamic_bs_stretch_package,
5580
onnx_execution_providers=["CUDAExecutionProvider", "CPUExecutionProvider"],
5681
)
57-
58-
# when
5982
predictions = model(
6083
[coins_counting_image_numpy, coins_counting_image_numpy],
6184
confidence=0.25,
6285
iou_threshold=0.45,
6386
max_detections=100,
6487
)
6588

66-
# then
67-
assert isinstance(predictions, list)
6889
assert len(predictions) == 2
6990
for pred in predictions:
7091
assert pred.xyxy.shape[1] == 4
71-
assert pred.xyxy.dtype == torch.int32
7292
assert len(pred.confidence) > 0
7393
assert torch.all(pred.confidence >= 0.25)
74-
assert torch.all(pred.confidence <= 1.0)
75-
# Both images are identical, so detections should match
7694
assert predictions[0].xyxy.shape == predictions[1].xyxy.shape
7795
assert torch.allclose(
7896
predictions[0].confidence, predictions[1].confidence, atol=0.01
@@ -81,98 +99,145 @@ def test_onnx_package_with_dynamic_batch_size_and_letterbox_batch_numpy(
8199

82100
@pytest.mark.slow
83101
@pytest.mark.onnx_extras
84-
def test_onnx_package_with_dynamic_batch_size_and_letterbox_torch(
85-
coin_counting_yololite_n_onnx_dynamic_bs_letterbox_package: str,
102+
def test_dynamic_non_fused_torch(
103+
coin_counting_yololite_edge_n_onnx_dynamic_bs_stretch_package: str,
86104
coins_counting_image_torch: torch.Tensor,
87105
) -> None:
88-
# given
89106
from inference_models.models.yololite.yololite_object_detection_onnx import (
90107
YOLOLiteForObjectDetectionOnnx,
91108
)
92109

93110
model = YOLOLiteForObjectDetectionOnnx.from_pretrained(
94-
model_name_or_path=coin_counting_yololite_n_onnx_dynamic_bs_letterbox_package,
111+
model_name_or_path=coin_counting_yololite_edge_n_onnx_dynamic_bs_stretch_package,
95112
onnx_execution_providers=["CUDAExecutionProvider", "CPUExecutionProvider"],
96113
)
97-
98-
# when
99114
predictions = model(
100-
coins_counting_image_torch,
101-
confidence=0.25,
102-
iou_threshold=0.45,
103-
max_detections=100,
115+
coins_counting_image_torch, confidence=0.25, iou_threshold=0.45
104116
)
105117

106-
# then
107-
assert isinstance(predictions, list)
108118
assert len(predictions) == 1
109119
assert predictions[0].xyxy.shape[1] == 4
110-
assert predictions[0].xyxy.dtype == torch.int32
111120
assert len(predictions[0].confidence) > 0
112121
assert torch.all(predictions[0].confidence >= 0.25)
113122

114123

115124
@pytest.mark.slow
116125
@pytest.mark.onnx_extras
117-
def test_onnx_high_confidence_threshold_returns_fewer_detections(
118-
coin_counting_yololite_n_onnx_dynamic_bs_letterbox_package: str,
126+
def test_dynamic_non_fused_high_confidence_returns_fewer(
127+
coin_counting_yololite_edge_n_onnx_dynamic_bs_stretch_package: str,
119128
coins_counting_image_numpy: np.ndarray,
120129
) -> None:
121-
# given
122130
from inference_models.models.yololite.yololite_object_detection_onnx import (
123131
YOLOLiteForObjectDetectionOnnx,
124132
)
125133

126134
model = YOLOLiteForObjectDetectionOnnx.from_pretrained(
127-
model_name_or_path=coin_counting_yololite_n_onnx_dynamic_bs_letterbox_package,
135+
model_name_or_path=coin_counting_yololite_edge_n_onnx_dynamic_bs_stretch_package,
128136
onnx_execution_providers=["CUDAExecutionProvider", "CPUExecutionProvider"],
129137
)
138+
low = model(coins_counting_image_numpy, confidence=0.1, iou_threshold=0.45)
139+
high = model(coins_counting_image_numpy, confidence=0.8, iou_threshold=0.45)
140+
141+
assert len(low[0].confidence) >= len(high[0].confidence)
142+
143+
144+
@pytest.mark.slow
145+
@pytest.mark.onnx_extras
146+
def test_dynamic_non_fused_class_agnostic_nms(
147+
coin_counting_yololite_edge_n_onnx_dynamic_bs_stretch_package: str,
148+
coins_counting_image_numpy: np.ndarray,
149+
) -> None:
150+
from inference_models.models.yololite.yololite_object_detection_onnx import (
151+
YOLOLiteForObjectDetectionOnnx,
152+
)
130153

131-
# when
132-
low_conf_predictions = model(
133-
coins_counting_image_numpy, confidence=0.1, iou_threshold=0.45
154+
model = YOLOLiteForObjectDetectionOnnx.from_pretrained(
155+
model_name_or_path=coin_counting_yololite_edge_n_onnx_dynamic_bs_stretch_package,
156+
onnx_execution_providers=["CUDAExecutionProvider", "CPUExecutionProvider"],
134157
)
135-
high_conf_predictions = model(
136-
coins_counting_image_numpy, confidence=0.8, iou_threshold=0.45
158+
standard = model(
159+
coins_counting_image_numpy, confidence=0.25, iou_threshold=0.45,
160+
class_agnostic_nms=False,
161+
)
162+
agnostic = model(
163+
coins_counting_image_numpy, confidence=0.25, iou_threshold=0.45,
164+
class_agnostic_nms=True,
137165
)
138166

139-
# then
140-
assert len(low_conf_predictions[0].confidence) >= len(
141-
high_conf_predictions[0].confidence
167+
assert len(agnostic[0].confidence) <= len(standard[0].confidence)
168+
169+
170+
# ── NMS-fused ────────────────────────────────────────────────────────────────
171+
172+
173+
@pytest.mark.slow
174+
@pytest.mark.onnx_extras
175+
def test_fused_nms_numpy(
176+
coin_counting_yololite_edge_n_onnx_dynamic_bs_stretch_fused_nms_package: str,
177+
coins_counting_image_numpy: np.ndarray,
178+
) -> None:
179+
from inference_models.models.yololite.yololite_object_detection_onnx import (
180+
YOLOLiteForObjectDetectionOnnx,
181+
)
182+
183+
model = YOLOLiteForObjectDetectionOnnx.from_pretrained(
184+
model_name_or_path=coin_counting_yololite_edge_n_onnx_dynamic_bs_stretch_fused_nms_package,
185+
onnx_execution_providers=["CUDAExecutionProvider", "CPUExecutionProvider"],
142186
)
187+
predictions = model(coins_counting_image_numpy, confidence=0.25)
188+
189+
assert len(predictions) == 1
190+
assert predictions[0].xyxy.shape[1] == 4
191+
assert predictions[0].xyxy.dtype == torch.int32
192+
assert len(predictions[0].confidence) > 0
193+
assert torch.all(predictions[0].confidence >= 0.25)
194+
assert torch.all(predictions[0].confidence <= 1.0)
143195

144196

145197
@pytest.mark.slow
146198
@pytest.mark.onnx_extras
147-
def test_onnx_class_agnostic_nms(
148-
coin_counting_yololite_n_onnx_dynamic_bs_letterbox_package: str,
199+
def test_fused_nms_batch_numpy(
200+
coin_counting_yololite_edge_n_onnx_dynamic_bs_stretch_fused_nms_package: str,
149201
coins_counting_image_numpy: np.ndarray,
150202
) -> None:
151-
# given
152203
from inference_models.models.yololite.yololite_object_detection_onnx import (
153204
YOLOLiteForObjectDetectionOnnx,
154205
)
155206

156207
model = YOLOLiteForObjectDetectionOnnx.from_pretrained(
157-
model_name_or_path=coin_counting_yololite_n_onnx_dynamic_bs_letterbox_package,
208+
model_name_or_path=coin_counting_yololite_edge_n_onnx_dynamic_bs_stretch_fused_nms_package,
158209
onnx_execution_providers=["CUDAExecutionProvider", "CPUExecutionProvider"],
159210
)
211+
predictions = model(
212+
[coins_counting_image_numpy, coins_counting_image_numpy], confidence=0.25,
213+
)
160214

161-
# when
162-
standard_predictions = model(
163-
coins_counting_image_numpy,
164-
confidence=0.25,
165-
iou_threshold=0.45,
166-
class_agnostic_nms=False,
215+
assert len(predictions) == 2
216+
for pred in predictions:
217+
assert pred.xyxy.shape[1] == 4
218+
assert len(pred.confidence) > 0
219+
assert torch.all(pred.confidence >= 0.25)
220+
assert predictions[0].xyxy.shape == predictions[1].xyxy.shape
221+
assert torch.allclose(
222+
predictions[0].confidence, predictions[1].confidence, atol=0.01
167223
)
168-
agnostic_predictions = model(
169-
coins_counting_image_numpy,
170-
confidence=0.25,
171-
iou_threshold=0.45,
172-
class_agnostic_nms=True,
224+
225+
226+
@pytest.mark.slow
227+
@pytest.mark.onnx_extras
228+
def test_fused_nms_high_confidence_returns_fewer(
229+
coin_counting_yololite_edge_n_onnx_dynamic_bs_stretch_fused_nms_package: str,
230+
coins_counting_image_numpy: np.ndarray,
231+
) -> None:
232+
from inference_models.models.yololite.yololite_object_detection_onnx import (
233+
YOLOLiteForObjectDetectionOnnx,
173234
)
174235

175-
# then - class-agnostic NMS should suppress more overlapping boxes
176-
assert len(agnostic_predictions[0].confidence) <= len(
177-
standard_predictions[0].confidence
236+
model = YOLOLiteForObjectDetectionOnnx.from_pretrained(
237+
model_name_or_path=coin_counting_yololite_edge_n_onnx_dynamic_bs_stretch_fused_nms_package,
238+
onnx_execution_providers=["CUDAExecutionProvider", "CPUExecutionProvider"],
178239
)
240+
low = model(coins_counting_image_numpy, confidence=0.1)
241+
high = model(coins_counting_image_numpy, confidence=0.8)
242+
243+
assert len(low[0].confidence) >= len(high[0].confidence)

0 commit comments

Comments
 (0)