diff --git a/README.md b/README.md index 9c585e3..4b9e802 100644 --- a/README.md +++ b/README.md @@ -1,12 +1,11 @@ -！！！！因为ncnn onnx2ncnn的bug导致crop算子计算有问题，故模型的指标需重新测试:https://github.com/Tencent/ncnn/pull/3999 +！！！！因为ncnn onnx2ncnn的bug导致crop算子计算有问题，修复后模型的指标作了调整修正:https://github.com/Tencent/ncnn/pull/3999 -！！！！Because the bug of ncnn onnx2ncnn causes problems in the calculation of the crop operator, the indicators of the model need to be retested :https://github.com/Tencent/ncnn/pull/3999 # :zap:FastestDet:zap:[![DOI](https://zenodo.org/badge/508635170.svg)](https://zenodo.org/badge/latestdoi/508635170) ![image](https://github.com/dog-qiuqiu/FastestDet/blob/main/data/data.png) * ***Faster! Stronger! Simpler!*** -* ***It has better single core reasoning performance and simpler feature map post-processing than Yolo-fastest*** -* ***In the ARM CPU of RK3568, the single core reasoning performance is 50% higher than Yolo-fastest*** -* ***The coco evaluation index increased by 3.8% compared with the map0.5 of Yolo-fastest*** +* ***It has better performance and simpler feature map post-processing than Yolo-fastest*** +* ***The performance is 10% higher than Yolo-fastest*** +* ***The coco evaluation index increased by 0.9% compared with the map0.5 of Yolo-fastestv2*** * ***算法介绍：https://zhuanlan.zhihu.com/p/536500269 交流qq群:1062122604*** # Evaluating indicator/Benchmark Network|mAPval 0.5|mAPval 0.5:0.95|Resolution|Run Time(4xCore)|Run Time(1xCore)|Params(M) @@ -17,7 +16,7 @@ Network|mAPval 0.5|mAPval 0.5:0.95|Resolution|Run Time(4xCore)|Run Time(1xCore)| [nanodet_m](https://github.com/RangiLyu/nanodet)|-|20.6%|320X320|49.24ms|160.35ms|0.95M [yolo-fastestv1.1](https://github.com/dog-qiuqiu/Yolo-Fastest/tree/master/ModelZoo/yolo-fastest-1.1_coco)|24.40%|-|320X320|26.60ms|75.74ms|0.35M [yolo-fastestv2](https://github.com/dog-qiuqiu/Yolo-FastestV2/tree/main/modelzoo)|24.10%|-|352X352|23.8ms|68.9ms|0.25M -FastestDet|27.8%|14.0%|512X512|21.51ms|34.62ms|0.25M +FastestDet|25.0%|12.3%|352X352|23.51ms|70.62ms|0.24M * ***Test platform Radxa Rock3A RK3568 ARM Cortex-A55 CPU，Based on [NCNN](https://github.com/Tencent/ncnn)*** * ***CPU lock frequency 2.0GHz*** # Improvement @@ -28,10 +27,9 @@ FastestDet|27.8%|14.0%|512X512|21.51ms|34.62ms|0.25M # Multi-platform benchmark Equipment|Computing backend|System|Framework|Run time(Single core)|Run time(Multi core) :---:|:---:|:---:|:---:|:---:|:---: -Radxa rock3a|RK3568(arm-cpu)|Linux(aarch64)|ncnn|34.62ms|21.51ms -Qualcomm|Snapdragon 835(arm-cpu)|Android(aarch64)|ncnn|28.34ms|16.24ms -AMD|R5-5600(X86-cpu)|Linux(amd64)|ncnn|2.16ms|1.73ms -Intel|i7-8700(X86-cpu)|Linux(amd64)|ncnn|5.21ms|4.73ms +Radxa rock3a|RK3568(arm-cpu)|Linux(aarch64)|ncnn|70.62ms|23.51ms +Qualcomm|Snapdragon 835(arm-cpu)|Android(aarch64)|ncnn|32.34ms|16.24ms +Intel|i7-8700(X86-cpu)|Linux(amd64)|ncnn|4.51ms|4.33ms # How to use ## Dependent installation * PiP(Note pytorch CUDA version selection) @@ -41,7 +39,7 @@ Intel|i7-8700(X86-cpu)|Linux(amd64)|ncnn|5.21ms|4.73ms ## Test * Picture test ``` - python3 test.py --yaml configs/coco.yaml --weight weights/weight_AP05\:0.278_280-epoch.pth --img data/3.jpg + python3 test.py --yaml configs/coco.yaml --weight weights/coco_ap05_0.250_280epoch.pth --img data/3.jpg ```
/> @@ -128,8 +126,8 @@ DATASET: NAMES: "dataset/coco128/coco.names" # .names category label file MODEL: NC: 80 # Number of detection categories - INPUT_WIDTH: 512 # The width of the model input image - INPUT_HEIGHT: 512 # The height of the model input image + INPUT_WIDTH: 352 # The width of the model input image + INPUT_HEIGHT: 352 # The height of the model input image TRAIN: LR: 0.001 # Train learn rate THRESH: 0.25 # ？？？？ @@ -149,7 +147,7 @@ TRAIN: ### Evaluation * Calculate map evaluation ``` - python3 eval.py --yaml configs/coco.yaml --weight weights/weight_AP05\:0.278_280-epoch.pth + python3 eval.py --yaml configs/coco.yaml --weight weights/coco_ap05_0.250_280epoch.pth ``` * COCO2017 evaluation ``` @@ -162,25 +160,25 @@ TRAIN: DONE (t=30.85s). Accumulating evaluation results... DONE (t=4.97s). - Average Precision (AP) @[ IoU=0.50:0.95 | area= all | maxDets=100 ] = 0.140 - Average Precision (AP) @[ IoU=0.50 | area= all | maxDets=100 ] = 0.278 - Average Precision (AP) @[ IoU=0.75 | area= all | maxDets=100 ] = 0.128 - Average Precision (AP) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.018 - Average Precision (AP) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.103 - Average Precision (AP) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.232 - Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets= 1 ] = 0.157 - Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets= 10 ] = 0.225 - Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets=100 ] = 0.231 - Average Recall (AR) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.032 - Average Recall (AR) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.201 - Average Recall (AR) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.359 + Average Precision (AP) @[ IoU=0.50:0.95 | area= all | maxDets=100 ] = 0.123 + Average Precision (AP) @[ IoU=0.50 | area= all | maxDets=100 ] = 0.250 + Average Precision (AP) @[ IoU=0.75 | area= all | maxDets=100 ] = 0.109 + Average Precision (AP) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.017 + Average Precision (AP) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.115 + Average Precision (AP) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.238 + Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets= 1 ] = 0.139 + Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets= 10 ] = 0.199 + Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets=100 ] = 0.205 + Average Recall (AR) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.035 + Average Recall (AR) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.218 + Average Recall (AR) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.374 ``` # Deploy ## Export onnx * You can export .onnx by adding the --onnx option when executing test.py ``` - python3 test.py --yaml configs/coco.yaml --weight weights/weight_AP05\:0.278_280-epoch.pth --img data/3.jpg --onnx + python3 test.py --yaml configs/coco.yaml --weight weights/coco_ap05_0.250_280epoch.pth --img data/3.jpg --onnx ``` ## onnx-runtime * You can learn about the pre and post-processing methods of FastestDet in this Sample diff --git a/configs/coco.yaml b/configs/coco.yaml index 1fd789a..373c637 100644 --- a/configs/coco.yaml +++ b/configs/coco.yaml @@ -4,15 +4,15 @@ DATASET: NAMES: "configs/coco.names" MODEL: NC: 80 - INPUT_WIDTH: 512 - INPUT_HEIGHT: 512 + INPUT_WIDTH: 352 + INPUT_HEIGHT: 352 TRAIN: LR: 0.001 THRESH: 0.25 WARMUP: true - BATCH_SIZE: 64 - END_EPOCH: 350 + BATCH_SIZE: 96 + END_EPOCH: 300 MILESTIONES: - - 150 - - 250 - - 300 \ No newline at end of file + - 100 + - 200 + - 250 \ No newline at end of file diff --git a/data/0_result.png b/data/0_result.png index 6969e89..972bc86 100644 Binary files a/data/0_result.png and b/data/0_result.png differ diff --git a/data/1_result.png b/data/1_result.png index e35cda6..541ed5e 100644 Binary files a/data/1_result.png and b/data/1_result.png differ diff --git a/data/2_result.png b/data/2_result.png index 4928116..8468fe8 100644 Binary files a/data/2_result.png and b/data/2_result.png differ diff --git a/data/3_result.png b/data/3_result.png index fa73b73..5b11514 100644 Binary files a/data/3_result.png and b/data/3_result.png differ diff --git a/data/4_result.png b/data/4_result.png index 853ac77..7585b41 100644 Binary files a/data/4_result.png and b/data/4_result.png differ diff --git a/example/onnx-runtime/FastestDet.onnx b/example/onnx-runtime/FastestDet.onnx index f82f538..0666a12 100644 Binary files a/example/onnx-runtime/FastestDet.onnx and b/example/onnx-runtime/FastestDet.onnx differ diff --git a/example/onnx-runtime/result.jpg b/example/onnx-runtime/result.jpg index 04d02e1..5cae23d 100644 Binary files a/example/onnx-runtime/result.jpg and b/example/onnx-runtime/result.jpg differ diff --git a/example/onnx-runtime/runtime.py b/example/onnx-runtime/runtime.py index 12979d0..75a8eaa 100755 --- a/example/onnx-runtime/runtime.py +++ b/example/onnx-runtime/runtime.py @@ -116,7 +116,7 @@ if __name__ == '__main__': # 读取图片 img = cv2.imread("1.jpg") # 模型输入的宽高 - input_width, input_height = 512, 512 + input_width, input_height = 352, 352 # 加载模型 session = onnxruntime.InferenceSession('FastestDet.onnx') # 目标检测 diff --git a/result.png b/result.png index fa73b73..5b11514 100644 Binary files a/result.png and b/result.png differ diff --git a/test.py b/test.py index d623f4b..4f5671f 100644 --- a/test.py +++ b/test.py @@ -13,7 +13,7 @@ if __name__ == '__main__': parser.add_argument('--yaml', type=str, default="", help='.yaml config') parser.add_argument('--weight', type=str, default=None, help='.weight config') parser.add_argument('--img', type=str, default='', help='The path of test image') - parser.add_argument('--thresh', type=float, default=0.8, help='The path of test image') + parser.add_argument('--thresh', type=float, default=0.65, help='The path of test image') parser.add_argument('--onnx', action="store_true", default=False, help='Export onnx file') parser.add_argument('--cpu', action="store_true", default=False, help='Run on cpu') diff --git a/utils/tool.py b/utils/tool.py index 82380b0..021b71d 100644 --- a/utils/tool.py +++ b/utils/tool.py @@ -58,7 +58,7 @@ class EMA(): self.backup = {} # 后处理(归一化后的坐标) -def handel_preds(preds, device, conf_thresh=0.25, nms_thresh=0.35): +def handel_preds(preds, device, conf_thresh=0.25, nms_thresh=0.45): total_bboxes, output_bboxes = [], [] # 将特征图转换为检测框的坐标 N, C, H, W = preds.shape diff --git a/weights/coco_ap05_0.250_280epoch.pth b/weights/coco_ap05_0.250_280epoch.pth new file mode 100644 index 0000000..cd78427 Binary files /dev/null and b/weights/coco_ap05_0.250_280epoch.pth differ diff --git a/weights/weight_AP05:0.278_280-epoch.pth b/weights/weight_AP05:0.278_280-epoch.pth deleted file mode 100644 index 9733fc9..0000000 Binary files a/weights/weight_AP05:0.278_280-epoch.pth and /dev/null differ