faster-coco-eval 1.7.2__tar.gz → 1.8.1__tar.gz

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (76) hide show
  1. faster_coco_eval-1.8.1/MANIFEST.in +5 -0
  2. {faster_coco_eval-1.7.2/faster_coco_eval.egg-info → faster_coco_eval-1.8.1}/PKG-INFO +30 -28
  3. {faster_coco_eval-1.7.2 → faster_coco_eval-1.8.1}/README.md +27 -22
  4. {faster_coco_eval-1.7.2 → faster_coco_eval-1.8.1}/csrc/faster_eval_api/coco_eval/cocoeval.cpp +388 -140
  5. {faster_coco_eval-1.7.2 → faster_coco_eval-1.8.1}/csrc/faster_eval_api/coco_eval/dataset.cpp +214 -32
  6. {faster_coco_eval-1.7.2 → faster_coco_eval-1.8.1}/csrc/faster_eval_api/coco_eval/dataset.h +8 -2
  7. {faster_coco_eval-1.7.2 → faster_coco_eval-1.8.1}/csrc/faster_eval_api/faster_eval_api.cpp +3 -0
  8. {faster_coco_eval-1.7.2 → faster_coco_eval-1.8.1}/csrc/mask_api/src/mask.cpp +212 -90
  9. {faster_coco_eval-1.7.2 → faster_coco_eval-1.8.1}/csrc/mask_api/src/mask.h +32 -2
  10. {faster_coco_eval-1.7.2 → faster_coco_eval-1.8.1}/csrc/mask_api/src/rle.cpp +34 -10
  11. {faster_coco_eval-1.7.2 → faster_coco_eval-1.8.1}/faster_coco_eval/core/coco.py +172 -84
  12. {faster_coco_eval-1.7.2 → faster_coco_eval-1.8.1}/faster_coco_eval/core/cocoeval.py +119 -88
  13. {faster_coco_eval-1.7.2 → faster_coco_eval-1.8.1}/faster_coco_eval/core/faster_eval_api.py +115 -23
  14. {faster_coco_eval-1.7.2 → faster_coco_eval-1.8.1}/faster_coco_eval/core/mask.py +12 -19
  15. {faster_coco_eval-1.7.2 → faster_coco_eval-1.8.1}/faster_coco_eval/extra/curves.py +40 -21
  16. {faster_coco_eval-1.7.2 → faster_coco_eval-1.8.1}/faster_coco_eval/extra/display.py +14 -13
  17. {faster_coco_eval-1.7.2 → faster_coco_eval-1.8.1}/faster_coco_eval/extra/draw.py +49 -34
  18. {faster_coco_eval-1.7.2 → faster_coco_eval-1.8.1}/faster_coco_eval/extra/extra.py +21 -22
  19. {faster_coco_eval-1.7.2 → faster_coco_eval-1.8.1}/faster_coco_eval/extra/utils.py +4 -1
  20. {faster_coco_eval-1.7.2 → faster_coco_eval-1.8.1}/faster_coco_eval/utils/pytorch/coco_dataset.py +7 -10
  21. {faster_coco_eval-1.7.2 → faster_coco_eval-1.8.1}/faster_coco_eval/utils/pytorch/coco_eval.py +19 -22
  22. faster_coco_eval-1.8.1/faster_coco_eval/version.py +2 -0
  23. {faster_coco_eval-1.7.2 → faster_coco_eval-1.8.1/faster_coco_eval.egg-info}/PKG-INFO +30 -28
  24. {faster_coco_eval-1.7.2 → faster_coco_eval-1.8.1}/faster_coco_eval.egg-info/SOURCES.txt +13 -1
  25. faster_coco_eval-1.8.1/history.md +228 -0
  26. {faster_coco_eval-1.7.2 → faster_coco_eval-1.8.1}/pyproject.toml +16 -4
  27. faster_coco_eval-1.8.1/requirements/docs.txt +6 -0
  28. {faster_coco_eval-1.7.2 → faster_coco_eval-1.8.1}/setup.py +5 -6
  29. faster_coco_eval-1.8.1/tests/test_annotation_id_zero.py +56 -0
  30. {faster_coco_eval-1.7.2 → faster_coco_eval-1.8.1}/tests/test_basic.py +45 -20
  31. {faster_coco_eval-1.7.2 → faster_coco_eval-1.8.1}/tests/test_boundary.py +5 -1
  32. faster_coco_eval-1.8.1/tests/test_coco_api_and_evaluator_regressions.py +705 -0
  33. {faster_coco_eval-1.7.2 → faster_coco_eval-1.8.1}/tests/test_coco_metric.py +3 -1
  34. faster_coco_eval-1.8.1/tests/test_compute_oks.py +69 -0
  35. faster_coco_eval-1.8.1/tests/test_core_low_regressions.py +105 -0
  36. faster_coco_eval-1.8.1/tests/test_cpp_safety_regressions.py +214 -0
  37. {faster_coco_eval-1.7.2 → faster_coco_eval-1.8.1}/tests/test_crowdpose.py +4 -1
  38. {faster_coco_eval-1.7.2 → faster_coco_eval-1.8.1}/tests/test_dataset.py +226 -0
  39. faster_coco_eval-1.8.1/tests/test_display_matrix_normalization.py +42 -0
  40. {faster_coco_eval-1.7.2 → faster_coco_eval-1.8.1}/tests/test_extensive_pycocotools_comparison.py +190 -168
  41. faster_coco_eval-1.8.1/tests/test_extra_curves.py +58 -0
  42. {faster_coco_eval-1.7.2 → faster_coco_eval-1.8.1}/tests/test_extra_draw.py +36 -8
  43. faster_coco_eval-1.8.1/tests/test_extra_regressions.py +92 -0
  44. {faster_coco_eval-1.7.2 → faster_coco_eval-1.8.1}/tests/test_extra_utils.py +42 -6
  45. {faster_coco_eval-1.7.2 → faster_coco_eval-1.8.1}/tests/test_keypoints.py +14 -4
  46. {faster_coco_eval-1.7.2 → faster_coco_eval-1.8.1}/tests/test_lvis_metric.py +38 -0
  47. {faster_coco_eval-1.7.2 → faster_coco_eval-1.8.1}/tests/test_mask_api.py +249 -0
  48. faster_coco_eval-1.8.1/tests/test_orphan_annotations.py +88 -0
  49. faster_coco_eval-1.8.1/tests/test_pytorch_coco_dataset.py +56 -0
  50. faster_coco_eval-1.8.1/tests/test_pytorch_coco_eval.py +68 -0
  51. {faster_coco_eval-1.7.2 → faster_coco_eval-1.8.1}/tests/test_simple_extra.py +93 -4
  52. {faster_coco_eval-1.7.2 → faster_coco_eval-1.8.1}/tests/test_torchmetrics.py +19 -8
  53. faster_coco_eval-1.8.1/tests/test_world_evaluator.py +124 -0
  54. faster_coco_eval-1.7.2/MANIFEST.in +0 -1
  55. faster_coco_eval-1.7.2/faster_coco_eval/version.py +0 -2
  56. faster_coco_eval-1.7.2/requirements/docs.txt +0 -8
  57. faster_coco_eval-1.7.2/tests/test_world_evalutor.py +0 -84
  58. {faster_coco_eval-1.7.2 → faster_coco_eval-1.8.1}/LICENSE +0 -0
  59. {faster_coco_eval-1.7.2 → faster_coco_eval-1.8.1}/csrc/faster_eval_api/coco_eval/cocoeval.h +0 -0
  60. {faster_coco_eval-1.7.2 → faster_coco_eval-1.8.1}/csrc/faster_eval_api/coco_eval/types.h +0 -0
  61. {faster_coco_eval-1.7.2 → faster_coco_eval-1.8.1}/csrc/mask_api/mask_api.cpp +0 -0
  62. {faster_coco_eval-1.7.2 → faster_coco_eval-1.8.1}/faster_coco_eval/__init__.py +0 -0
  63. {faster_coco_eval-1.7.2 → faster_coco_eval-1.8.1}/faster_coco_eval/core/__init__.py +0 -0
  64. {faster_coco_eval-1.7.2 → faster_coco_eval-1.8.1}/faster_coco_eval/extra/__init__.py +0 -0
  65. {faster_coco_eval-1.7.2 → faster_coco_eval-1.8.1}/faster_coco_eval/utils/__init__.py +0 -0
  66. {faster_coco_eval-1.7.2 → faster_coco_eval-1.8.1}/faster_coco_eval/utils/pytorch/__init__.py +0 -0
  67. {faster_coco_eval-1.7.2 → faster_coco_eval-1.8.1}/faster_coco_eval.egg-info/dependency_links.txt +0 -0
  68. {faster_coco_eval-1.7.2 → faster_coco_eval-1.8.1}/faster_coco_eval.egg-info/requires.txt +0 -0
  69. {faster_coco_eval-1.7.2 → faster_coco_eval-1.8.1}/faster_coco_eval.egg-info/top_level.txt +0 -0
  70. {faster_coco_eval-1.7.2 → faster_coco_eval-1.8.1}/requirements/extra.txt +0 -0
  71. {faster_coco_eval-1.7.2 → faster_coco_eval-1.8.1}/requirements/runtime.txt +0 -0
  72. {faster_coco_eval-1.7.2 → faster_coco_eval-1.8.1}/requirements/tests.txt +0 -0
  73. {faster_coco_eval-1.7.2 → faster_coco_eval-1.8.1}/setup.cfg +0 -0
  74. {faster_coco_eval-1.7.2 → faster_coco_eval-1.8.1}/tests/test_cocoapi_fake_data.py +0 -0
  75. {faster_coco_eval-1.7.2 → faster_coco_eval-1.8.1}/tests/test_init_pycocotools.py +0 -0
  76. {faster_coco_eval-1.7.2 → faster_coco_eval-1.8.1}/tests/test_ranges.py +0 -0
@@ -0,0 +1,5 @@
1
+ graft csrc
2
+ graft requirements
3
+ include README.md
4
+ include history.md
5
+ include LICENSE
@@ -1,22 +1,19 @@
1
1
  Metadata-Version: 2.4
2
2
  Name: faster-coco-eval
3
- Version: 1.7.2
3
+ Version: 1.8.1
4
4
  Summary: Faster interpretation of the original COCOEval
5
5
  Home-page: https://github.com/MiXaiLL76/faster_coco_eval
6
- Author: MiXaiLL76
6
+ Author: MiXaiLL76,Borda
7
7
  Author-email: mike.milos@yandex.ru
8
8
  Classifier: Development Status :: 4 - Beta
9
9
  Classifier: Operating System :: OS Independent
10
10
  Classifier: Programming Language :: Python :: 3
11
- Classifier: Programming Language :: Python :: 3.7
12
- Classifier: Programming Language :: Python :: 3.8
13
- Classifier: Programming Language :: Python :: 3.9
14
11
  Classifier: Programming Language :: Python :: 3.10
15
12
  Classifier: Programming Language :: Python :: 3.11
16
13
  Classifier: Programming Language :: Python :: 3.12
17
14
  Classifier: Programming Language :: Python :: 3.13
18
15
  Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
19
- Requires-Python: >=3.7
16
+ Requires-Python: >=3.10
20
17
  Description-Content-Type: text/markdown
21
18
  License-File: LICENSE
22
19
  Requires-Dist: numpy
@@ -63,20 +60,20 @@ Dynamic: summary
63
60
  | ------------------------------ | ------------------------------------------------------------------------------------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
64
61
  | **Support & Development** | Outdated and not actively maintained. Issues and incompatibilities arise with new releases. | Actively maintained, continuously evolving, and regularly updated with new features and bug fixes. |
65
62
  | **Transparency & Reliability** | Lacks comprehensive testing, making updates risky and results less predictable. | Emphasizes extensive test coverage and code quality, ensuring trustworthy and reliable results. |
66
- | **Performance** | Significantly slower, especially on large datasets or distributed workloads. | **3-4x faster** due to C++ optimizations and modern algorithms. |
63
+ | **Performance** | Significantly slower, especially on large datasets or distributed workloads. | Native C++ implementation with measured speedups that depend on the workload and hardware. |
67
64
  | **Functionality** | Limited to basic COCO format evaluation. | Offers extended metrics, support for new IoU types, compatibility with more datasets (e.g., CrowdPose, LVIS), advanced visualizations, and seamless integration with PyTorch/TorchVision. |
68
- | **Ease of Use** | Requires manual installation, often with compilation issues. | Simple `pip install`, no compilation required, and drop-in replacement API. |
69
- | **Visualization** | Basic plotting capabilities. | Advanced error visualization, annotation display, and comprehensive metric analysis tools. |
65
+ | **Ease of Use** | Requires manual installation, often with compilation issues. | Simple `pip install` with pre-built wheels for common CPython/glibc Linux, macOS, and Windows platforms; source builds may be needed elsewhere. |
66
+ | **Visualization** | Basic plotting capabilities. | Advanced error visualization, annotation display, and comprehensive metric analysis tools. |
70
67
 
71
- ---
68
+ ______________________________________________________________________
72
69
 
73
70
  **Key Benefits of Faster-COCO-Eval:**
74
71
 
75
- ✅ **Blazing Fast Performance** - Evaluate large datasets in minutes instead of hours
76
- ✅ **Reliable & Trusted** - Extensive test coverage ensures consistent, reproducible results
77
- ✅ **Modern Features** - Support for latest CV tasks, IoU types, and dataset formats
78
- ✅ **Easy to Use** - Drop-in replacement for pycocotools with enhanced API
79
- ✅ **Comprehensive Visualization** - Understand your model's performance with beautiful, informative plots
72
+ ✅ **Blazing Fast Performance** - Evaluate large datasets in minutes instead of hours
73
+ ✅ **Reliable & Trusted** - Extensive test coverage ensures consistent, reproducible results
74
+ ✅ **Modern Features** - Support for latest CV tasks, IoU types, and dataset formats
75
+ ✅ **Easy to Use** - Drop-in replacement for pycocotools with enhanced API
76
+ ✅ **Comprehensive Visualization** - Understand your model's performance with beautiful, informative plots
80
77
 
81
78
  **Join thousands of computer vision researchers and engineers who have already switched to Faster-COCO-Eval!**
82
79
 
@@ -121,6 +118,8 @@ from pycocotools.coco import COCO
121
118
  from pycocotools.cocoeval import COCOeval
122
119
 
123
120
  # Load annotations and predictions
121
+ anno_json = "path/to/annotations.json"
122
+ pred_json = "path/to/predictions.json"
124
123
  anno = COCO(str(anno_json)) # Annotations file
125
124
  pred = anno.loadRes(str(pred_json)) # Predictions file
126
125
 
@@ -137,22 +136,21 @@ val.accumulate()
137
136
  val.summarize()
138
137
  ```
139
138
 
140
- **That's it! Your existing code will run 3-4x faster with no changes.**
139
+ **That's it! Your existing code runs through the native evaluator with no API changes.**
141
140
 
142
141
  ## ⚡ Blazing Fast Performance
143
142
 
144
- Faster-COCO-Eval is built on top of a highly optimized C++ implementation, providing **3-4x faster evaluation** than the standard pycocotools.
143
+ Faster-COCO-Eval uses a native C++ implementation. The measured result below is one local reference point; real-world speedup varies with dataset size, metric, and hardware.
145
144
 
146
145
  ### Real-World Performance Benchmark
147
146
 
148
- Tested on 5000 images from the COCO validation dataset using mmdetection framework:
147
+ Local reference measurement: 100 synthetic images, 15 ground-truth boxes and 100 detections per image, five categories, bbox evaluation, seven timed samples after two warmups. The measurement was run on 2026-08-23 on macOS 26.6.1 arm64, Python 3.10.11, and NumPy 2.2.6. The implementations' `eval["precision"]` arrays matched within the comparison tolerance.
149
148
 
150
- | Evaluation Type | Faster-COCO-Eval (sec) | pycocotools (sec) | Speedup |
151
- |-----------------|------------------------|-------------------|---------|
152
- | Bounding Boxes | 5.812 | 22.72 | **3.9x** |
153
- | Segmentation | 7.413 | 24.434 | **3.3x** |
149
+ | Evaluation Type | Faster-COCO-Eval (sec) | pycocotools (sec) | Speedup |
150
+ | --------------- | ----------------------- | ----------------------- | --------- |
151
+ | Bounding Boxes | 0.053977 (MAD 0.001883) | 0.296960 (MAD 0.001667) | **5.50x** |
154
152
 
155
- **For large datasets, this means hours saved on evaluation time!**
153
+ This local measurement is hardware- and workload-dependent; it is not a universal performance guarantee.
156
154
 
157
155
  ### Colab Examples
158
156
 
@@ -166,6 +164,7 @@ See the performance in action:
166
164
  Faster-COCO-Eval goes beyond basic evaluation with these advanced capabilities:
167
165
 
168
166
  ### Core Evaluation
167
+
169
168
  - **Drop-in pycocotools replacement** - No code changes needed
170
169
  - **Support for all COCO metric types**: bbox, segm, keypoints
171
170
  - **LVIS (Large Vocabulary Instance Segmentation) evaluation**
@@ -173,6 +172,7 @@ Faster-COCO-Eval goes beyond basic evaluation with these advanced capabilities:
173
172
  - **Multiple IoU types**: standard, rotated, and custom IoU definitions
174
173
 
175
174
  ### Advanced Visualization
175
+
176
176
  - **Error visualization**: See where your model is making mistakes
177
177
  - **Annotation display**: Visualize ground truth and predictions together
178
178
  - **Metric curves**: Precision-recall curves, class-wise performance
@@ -180,12 +180,14 @@ Faster-COCO-Eval goes beyond basic evaluation with these advanced capabilities:
180
180
  - **Interactive Jupyter notebook examples**
181
181
 
182
182
  ### Modern Integrations
183
+
183
184
  - **PyTorch/TorchVision compatibility**
184
- - **Seamless integration with mmdetection, Detectron2, and YOLO frameworks**
185
+ - **Seamless integration with mmdetection and YOLO frameworks**
185
186
  - **Distributed evaluation support**
186
187
  - **Memory optimized for large datasets**
187
188
 
188
189
  ### Additional Tools
190
+
189
191
  - **Boundary evaluation for segmentation tasks**
190
192
  - **Custom dataset support**
191
193
  - **Comprehensive API documentation**
@@ -198,13 +200,13 @@ Faster-COCO-Eval prioritizes **correctness and reliability** through extensive t
198
200
  ### Comprehensive Test Suite
199
201
 
200
202
  - **90+ automated tests** covering all functionality
201
- - **Exact equality validation** against pycocotools across all metrics
202
- - **Continuous integration** on Python 3.9-3.13
203
+ - **Numerical parity checks** against pycocotools across the supported metrics
204
+ - **Continuous integration** on Python 3.10-3.13
203
205
  - **Edge case coverage** including boundary conditions and error handling
204
206
 
205
207
  ### Extensive PyCocoTools Comparison
206
208
 
207
- New comprehensive tests validate **exact numerical equality** with pycocotools:
209
+ New comprehensive tests validate **numerical parity within a strict tolerance** with pycocotools:
208
210
 
209
211
  - **Object Detection**: Tests with 10-100 images, hundreds to thousands of annotations
210
212
  - **Instance Segmentation**: RLE mask encoding and pixel-level IoU validation
@@ -212,7 +214,7 @@ New comprehensive tests validate **exact numerical equality** with pycocotools:
212
214
  - **Multiple Scenarios**: Small/medium/large objects, various confidence distributions
213
215
  - **Edge Cases**: Perfect predictions, low-confidence detections, mixed object sizes
214
216
 
215
- All tests confirm **bit-for-bit identical results** between faster_coco_eval and pycocotools, giving you confidence to use this library as a drop-in replacement while gaining 3-4x performance improvements.
217
+ The comparison suite checks numerical parity between faster_coco_eval and pycocotools, giving you confidence to use this library as a drop-in replacement. See the benchmark above for one measured bbox workload; it is not a universal performance guarantee.
216
218
 
217
219
  See [tests/README.md](tests/README.md) for detailed test documentation.
218
220
 
@@ -16,20 +16,20 @@
16
16
  | ------------------------------ | ------------------------------------------------------------------------------------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
17
17
  | **Support & Development** | Outdated and not actively maintained. Issues and incompatibilities arise with new releases. | Actively maintained, continuously evolving, and regularly updated with new features and bug fixes. |
18
18
  | **Transparency & Reliability** | Lacks comprehensive testing, making updates risky and results less predictable. | Emphasizes extensive test coverage and code quality, ensuring trustworthy and reliable results. |
19
- | **Performance** | Significantly slower, especially on large datasets or distributed workloads. | **3-4x faster** due to C++ optimizations and modern algorithms. |
19
+ | **Performance** | Significantly slower, especially on large datasets or distributed workloads. | Native C++ implementation with measured speedups that depend on the workload and hardware. |
20
20
  | **Functionality** | Limited to basic COCO format evaluation. | Offers extended metrics, support for new IoU types, compatibility with more datasets (e.g., CrowdPose, LVIS), advanced visualizations, and seamless integration with PyTorch/TorchVision. |
21
- | **Ease of Use** | Requires manual installation, often with compilation issues. | Simple `pip install`, no compilation required, and drop-in replacement API. |
22
- | **Visualization** | Basic plotting capabilities. | Advanced error visualization, annotation display, and comprehensive metric analysis tools. |
21
+ | **Ease of Use** | Requires manual installation, often with compilation issues. | Simple `pip install` with pre-built wheels for common CPython/glibc Linux, macOS, and Windows platforms; source builds may be needed elsewhere. |
22
+ | **Visualization** | Basic plotting capabilities. | Advanced error visualization, annotation display, and comprehensive metric analysis tools. |
23
23
 
24
- ---
24
+ ______________________________________________________________________
25
25
 
26
26
  **Key Benefits of Faster-COCO-Eval:**
27
27
 
28
- ✅ **Blazing Fast Performance** - Evaluate large datasets in minutes instead of hours
29
- ✅ **Reliable & Trusted** - Extensive test coverage ensures consistent, reproducible results
30
- ✅ **Modern Features** - Support for latest CV tasks, IoU types, and dataset formats
31
- ✅ **Easy to Use** - Drop-in replacement for pycocotools with enhanced API
32
- ✅ **Comprehensive Visualization** - Understand your model's performance with beautiful, informative plots
28
+ ✅ **Blazing Fast Performance** - Evaluate large datasets in minutes instead of hours
29
+ ✅ **Reliable & Trusted** - Extensive test coverage ensures consistent, reproducible results
30
+ ✅ **Modern Features** - Support for latest CV tasks, IoU types, and dataset formats
31
+ ✅ **Easy to Use** - Drop-in replacement for pycocotools with enhanced API
32
+ ✅ **Comprehensive Visualization** - Understand your model's performance with beautiful, informative plots
33
33
 
34
34
  **Join thousands of computer vision researchers and engineers who have already switched to Faster-COCO-Eval!**
35
35
 
@@ -74,6 +74,8 @@ from pycocotools.coco import COCO
74
74
  from pycocotools.cocoeval import COCOeval
75
75
 
76
76
  # Load annotations and predictions
77
+ anno_json = "path/to/annotations.json"
78
+ pred_json = "path/to/predictions.json"
77
79
  anno = COCO(str(anno_json)) # Annotations file
78
80
  pred = anno.loadRes(str(pred_json)) # Predictions file
79
81
 
@@ -90,22 +92,21 @@ val.accumulate()
90
92
  val.summarize()
91
93
  ```
92
94
 
93
- **That's it! Your existing code will run 3-4x faster with no changes.**
95
+ **That's it! Your existing code runs through the native evaluator with no API changes.**
94
96
 
95
97
  ## ⚡ Blazing Fast Performance
96
98
 
97
- Faster-COCO-Eval is built on top of a highly optimized C++ implementation, providing **3-4x faster evaluation** than the standard pycocotools.
99
+ Faster-COCO-Eval uses a native C++ implementation. The measured result below is one local reference point; real-world speedup varies with dataset size, metric, and hardware.
98
100
 
99
101
  ### Real-World Performance Benchmark
100
102
 
101
- Tested on 5000 images from the COCO validation dataset using mmdetection framework:
103
+ Local reference measurement: 100 synthetic images, 15 ground-truth boxes and 100 detections per image, five categories, bbox evaluation, seven timed samples after two warmups. The measurement was run on 2026-08-23 on macOS 26.6.1 arm64, Python 3.10.11, and NumPy 2.2.6. The implementations' `eval["precision"]` arrays matched within the comparison tolerance.
102
104
 
103
- | Evaluation Type | Faster-COCO-Eval (sec) | pycocotools (sec) | Speedup |
104
- |-----------------|------------------------|-------------------|---------|
105
- | Bounding Boxes | 5.812 | 22.72 | **3.9x** |
106
- | Segmentation | 7.413 | 24.434 | **3.3x** |
105
+ | Evaluation Type | Faster-COCO-Eval (sec) | pycocotools (sec) | Speedup |
106
+ | --------------- | ----------------------- | ----------------------- | --------- |
107
+ | Bounding Boxes | 0.053977 (MAD 0.001883) | 0.296960 (MAD 0.001667) | **5.50x** |
107
108
 
108
- **For large datasets, this means hours saved on evaluation time!**
109
+ This local measurement is hardware- and workload-dependent; it is not a universal performance guarantee.
109
110
 
110
111
  ### Colab Examples
111
112
 
@@ -119,6 +120,7 @@ See the performance in action:
119
120
  Faster-COCO-Eval goes beyond basic evaluation with these advanced capabilities:
120
121
 
121
122
  ### Core Evaluation
123
+
122
124
  - **Drop-in pycocotools replacement** - No code changes needed
123
125
  - **Support for all COCO metric types**: bbox, segm, keypoints
124
126
  - **LVIS (Large Vocabulary Instance Segmentation) evaluation**
@@ -126,6 +128,7 @@ Faster-COCO-Eval goes beyond basic evaluation with these advanced capabilities:
126
128
  - **Multiple IoU types**: standard, rotated, and custom IoU definitions
127
129
 
128
130
  ### Advanced Visualization
131
+
129
132
  - **Error visualization**: See where your model is making mistakes
130
133
  - **Annotation display**: Visualize ground truth and predictions together
131
134
  - **Metric curves**: Precision-recall curves, class-wise performance
@@ -133,12 +136,14 @@ Faster-COCO-Eval goes beyond basic evaluation with these advanced capabilities:
133
136
  - **Interactive Jupyter notebook examples**
134
137
 
135
138
  ### Modern Integrations
139
+
136
140
  - **PyTorch/TorchVision compatibility**
137
- - **Seamless integration with mmdetection, Detectron2, and YOLO frameworks**
141
+ - **Seamless integration with mmdetection and YOLO frameworks**
138
142
  - **Distributed evaluation support**
139
143
  - **Memory optimized for large datasets**
140
144
 
141
145
  ### Additional Tools
146
+
142
147
  - **Boundary evaluation for segmentation tasks**
143
148
  - **Custom dataset support**
144
149
  - **Comprehensive API documentation**
@@ -151,13 +156,13 @@ Faster-COCO-Eval prioritizes **correctness and reliability** through extensive t
151
156
  ### Comprehensive Test Suite
152
157
 
153
158
  - **90+ automated tests** covering all functionality
154
- - **Exact equality validation** against pycocotools across all metrics
155
- - **Continuous integration** on Python 3.9-3.13
159
+ - **Numerical parity checks** against pycocotools across the supported metrics
160
+ - **Continuous integration** on Python 3.10-3.13
156
161
  - **Edge case coverage** including boundary conditions and error handling
157
162
 
158
163
  ### Extensive PyCocoTools Comparison
159
164
 
160
- New comprehensive tests validate **exact numerical equality** with pycocotools:
165
+ New comprehensive tests validate **numerical parity within a strict tolerance** with pycocotools:
161
166
 
162
167
  - **Object Detection**: Tests with 10-100 images, hundreds to thousands of annotations
163
168
  - **Instance Segmentation**: RLE mask encoding and pixel-level IoU validation
@@ -165,7 +170,7 @@ New comprehensive tests validate **exact numerical equality** with pycocotools:
165
170
  - **Multiple Scenarios**: Small/medium/large objects, various confidence distributions
166
171
  - **Edge Cases**: Perfect predictions, low-confidence detections, mixed object sizes
167
172
 
168
- All tests confirm **bit-for-bit identical results** between faster_coco_eval and pycocotools, giving you confidence to use this library as a drop-in replacement while gaining 3-4x performance improvements.
173
+ The comparison suite checks numerical parity between faster_coco_eval and pycocotools, giving you confidence to use this library as a drop-in replacement. See the benchmark above for one measured bbox workload; it is not a universal performance guarantee.
169
174
 
170
175
  See [tests/README.md](tests/README.md) for detailed test documentation.
171
176