add gpu performence bench mark

benchmark_grid.json

@@ -2,91 +2,91 @@
   {
     "backbone": "vgg16",
     "attention": "none",
-    "places": "backbone_high,desc_head",
-    "single_ms_mean": 351.6519069671631,
-    "single_ms_std": 1.8778125281542124,
-    "fpn_ms_mean": 719.3304697672526,
-    "fpn_ms_std": 3.949980966745213,
-    "runs": 3
+    "places": "backbone_high",
+    "single_ms_mean": 4.528331756591797,
+    "single_ms_std": 0.018315389112121477,
+    "fpn_ms_mean": 8.5052490234375,
+    "fpn_ms_std": 0.0024987359059474757,
+    "runs": 5
   },
   {
     "backbone": "vgg16",
     "attention": "se",
-    "places": "backbone_high,desc_head",
-    "single_ms_mean": 349.7585455576579,
-    "single_ms_std": 1.9950684383137551,
-    "fpn_ms_mean": 721.4130560557047,
-    "fpn_ms_std": 2.7448351792281374,
-    "runs": 3
+    "places": "backbone_high",
+    "single_ms_mean": 3.79791259765625,
+    "single_ms_std": 0.014929344228397397,
+    "fpn_ms_mean": 7.117033004760742,
+    "fpn_ms_std": 0.0039580356539625425,
+    "runs": 5
   },
   {
     "backbone": "vgg16",
     "attention": "cbam",
-    "places": "backbone_high,desc_head",
-    "single_ms_mean": 354.4490337371826,
-    "single_ms_std": 1.4903953036396786,
-    "fpn_ms_mean": 744.7629769643148,
-    "fpn_ms_std": 29.3233387791729,
-    "runs": 3
+    "places": "backbone_high",
+    "single_ms_mean": 3.7283897399902344,
+    "single_ms_std": 0.01896289713396852,
+    "fpn_ms_mean": 6.954669952392578,
+    "fpn_ms_std": 0.0946284511822057,
+    "runs": 5
   },
   {
     "backbone": "resnet34",
     "attention": "none",
-    "places": "backbone_high,desc_head",
-    "single_ms_mean": 90.98696708679199,
-    "single_ms_std": 0.41179110533866975,
-    "fpn_ms_mean": 117.2173023223877,
-    "fpn_ms_std": 0.40632490569423124,
-    "runs": 3
+    "places": "backbone_high",
+    "single_ms_mean": 2.3172378540039062,
+    "single_ms_std": 0.03704733205002756,
+    "fpn_ms_mean": 2.7330875396728516,
+    "fpn_ms_std": 0.006544318567008118,
+    "runs": 5
   },
   {
     "backbone": "resnet34",
     "attention": "se",
-    "places": "backbone_high,desc_head",
-    "single_ms_mean": 90.78375498453777,
-    "single_ms_std": 0.4705899743190883,
-    "fpn_ms_mean": 115.90576171875,
-    "fpn_ms_std": 1.3081578935341862,
-    "runs": 3
+    "places": "backbone_high",
+    "single_ms_mean": 2.3345470428466797,
+    "single_ms_std": 0.01149701754726714,
+    "fpn_ms_mean": 2.7266979217529297,
+    "fpn_ms_std": 0.0040167693497949,
+    "runs": 5
   },
   {
     "backbone": "resnet34",
     "attention": "cbam",
-    "places": "backbone_high,desc_head",
-    "single_ms_mean": 96.49538993835449,
-    "single_ms_std": 3.17170034860506,
-    "fpn_ms_mean": 111.08938852945964,
-    "fpn_ms_std": 1.0126843546619573,
-    "runs": 3
+    "places": "backbone_high",
+    "single_ms_mean": 2.4645328521728516,
+    "single_ms_std": 0.03573384703501215,
+    "fpn_ms_mean": 2.7351856231689453,
+    "fpn_ms_std": 0.004198875420141471,
+    "runs": 5
   },
   {
     "backbone": "efficientnet_b0",
     "attention": "none",
-    "places": "backbone_high,desc_head",
-    "single_ms_mean": 40.451606114705406,
-    "single_ms_std": 1.5293525027201111,
-    "fpn_ms_mean": 127.30161348978679,
-    "fpn_ms_std": 0.08508800981401025,
-    "runs": 3
+    "places": "backbone_high",
+    "single_ms_mean": 3.6920547485351562,
+    "single_ms_std": 0.06926683030174544,
+    "fpn_ms_mean": 4.38084602355957,
+    "fpn_ms_std": 0.021533091774855868,
+    "runs": 5
   },
   {
     "backbone": "efficientnet_b0",
     "attention": "se",
-    "places": "backbone_high,desc_head",
-    "single_ms_mean": 46.480417251586914,
-    "single_ms_std": 0.2622188910897682,
-    "fpn_ms_mean": 142.35099156697592,
-    "fpn_ms_std": 6.611047958580852,
-    "runs": 3
+    "places": "backbone_high",
+    "single_ms_mean": 3.7618160247802734,
+    "single_ms_std": 0.05971848107723002,
+    "fpn_ms_mean": 4.3704986572265625,
+    "fpn_ms_std": 0.02873211962906253,
+    "runs": 5
   },
   {
     "backbone": "efficientnet_b0",
     "attention": "cbam",
-    "places": "backbone_high,desc_head",
-    "single_ms_mean": 47.10610707600912,
-    "single_ms_std": 0.47150733957171853,
-    "fpn_ms_mean": 150.99199612935385,
-    "fpn_ms_std": 12.465987661773038,
-    "runs": 3
+    "places": "backbone_high",
+    "single_ms_mean": 3.9876937866210938,
+    "single_ms_std": 0.07599183707384338,
+    "fpn_ms_mean": 4.412364959716797,
+    "fpn_ms_std": 0.023552763127197434,
+    "runs": 5
   }
 ]

docs/description/Performance_Benchmark.md

@@ -64,6 +64,48 @@ PYTHONPATH=. uv run python tests/benchmark_grid.py \
 
 运行会同时输出控制台摘要并保存 JSON：`benchmark_grid.json`。
 
+## GPU 测试结果（A100）
+
+最后更新：2025-01-XX  
+设备：NVIDIA A100（CUDA）  
+输入：1×3×512×512 随机张量  
+重复次数：5（每组）  
+注意力放置位置：backbone_high
+
+> 说明：本测试在 A100 GPU 上进行，展示了不同骨干网络和注意力模块组合在 GPU 上的推理性能。
+
+### 结果汇总（ms）
+
+| Backbone           | Attention | Single Mean ± Std | FPN Mean ± Std |
+|--------------------|-----------|------------------:|---------------:|
+| vgg16              | none      | 4.53 ± 0.02       | 8.51 ± 0.002   |
+| vgg16              | se        | 3.80 ± 0.01       | 7.12 ± 0.004   |
+| vgg16              | cbam      | 3.73 ± 0.02       | 6.95 ± 0.09    |
+| resnet34           | none      | 2.32 ± 0.04       | 2.73 ± 0.007   |
+| resnet34           | se        | 2.33 ± 0.01       | 2.73 ± 0.004   |
+| resnet34           | cbam      | 2.46 ± 0.04       | 2.74 ± 0.004   |
+| efficientnet_b0    | none      | 3.69 ± 0.07       | 4.38 ± 0.02    |
+| efficientnet_b0    | se        | 3.76 ± 0.06       | 4.37 ± 0.03    |
+| efficientnet_b0    | cbam      | 3.99 ± 0.08       | 4.41 ± 0.02    |
+
+复现实验：
+
+```zsh
+PYTHONPATH=. uv run python tests/benchmark_grid.py \
+  --device cuda --image-size 512 --runs 5 \
+  --backbones vgg16 resnet34 efficientnet_b0 \
+  --attentions none se cbam \
+  --places backbone_high
+```
+
+### GPU 测试观察
+
+- **ResNet34 表现最佳**：在 GPU 上，ResNet34 在单尺度和 FPN 路径上都表现出色，单尺度约 2.3ms，FPN 约 2.7ms。
+- **VGG16 在 GPU 上仍有明显开销**：尽管在 GPU 上加速，VGG16 仍然是三种骨干中最慢的，单尺度约 3.7-4.5ms。
+- **EfficientNet-B0 表现中等**：在 GPU 上介于 VGG16 和 ResNet34 之间，单尺度约 3.7-4.0ms。
+- **注意力模块影响较小**：在 GPU 上，注意力模块（SE、CBAM）对性能的影响相对较小，FPN 路径上的差异尤其不明显。
+- **FPN 开销相对可控**：在 GPU 上，FPN 路径相比单尺度的额外开销较小，ResNet34 仅增加约 18%。
+
 ## 观察与解读
 - vgg16 明显最慢，FPN 额外的横向/上采样代价在 CPU 上更突出（>2×）。
 - resnet34 在单尺度上显著快于 vgg16，FPN 增幅较小（约 +25%）。