2022.7.14:Optimize loss, adopt IOU aware based on smooth L1, and the AP is significantly increased by 0.7

:zap:FastestDet:zap:

• Faster! Stronger! Simpler!
• 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 1.2% 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) :---:|:---:|:---:|:---:|:---:|:---:|:---: yolov5s|56.8%|37.4%|640X640|395.31ms|1139.16ms|7.2M yolov6n|-|30.8%|416X416|109.24ms|445.44ms|4.3M yolox-nano|-|25.8%|416X416|76.31ms|191.16ms|0.91M nanodet_m|-|20.6%|320X320|49.24ms|160.35ms|0.95M yolo-fastestv1.1|24.40%|-|320X320|26.60ms|75.74ms|0.35M yolo-fastestv2|24.10%|-|352X352|23.8ms|68.9ms|0.25M FastestDet|25.3%|13.0%|352X352|23.51ms|70.62ms|0.24M

• Test platform Radxa Rock3A RK3568 ARM Cortex-A55 CPU，Based on NCNN
• CPU lock frequency 2.0GHz

Improvement

• Anchor-Free
• Single scale detector head
• Cross grid multiple candidate targets
• Dynamic positive and negative sample allocation

Multi-platform benchmark

Equipment|Computing backend|System|Framework|Run time(Single core)|Run time(Multi core) :---:|:---:|:---:|:---:|:---:|:---: Radxa rock3a|RK3568(arm-cpu)|Linux(aarch64)|ncnn|70.62ms|23.51ms Radxa rock3a|RK3568(NPU)|Linux(aarch64)|rknn|28ms|- 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)

  pip install -r requirements.txt
  

Test

• Picture test

  python3 test.py --yaml configs/coco.yaml --weight weights/weight_AP05:0.253207_280-epoch.pth --img data/3.jpg
  

<div align=center> <img src="https://github.com/dog-qiuqiu/FastestDet/blob/main/result.png"> /> </div>

How to train

Building data sets(The dataset is constructed in the same way as darknet yolo)

• The format of the data set is the same as that of Darknet Yolo, Each image corresponds to a .txt label file. The label format is also based on Darknet Yolo's data set label format: "category cx cy wh", where category is the category subscript, cx, cy are the coordinates of the center point of the normalized label box, and w, h are the normalized label box The width and height, .txt label file content example as follows:

  11 0.344192634561 0.611 0.416430594901 0.262
  14 0.509915014164 0.51 0.974504249292 0.972
  

• The image and its corresponding label file have the same name and are stored in the same directory. The data file structure is as follows:

  .
  ├── train
  │   ├── 000001.jpg
  │   ├── 000001.txt
  │   ├── 000002.jpg
  │   ├── 000002.txt
  │   ├── 000003.jpg
  │   └── 000003.txt
  └── val
      ├── 000043.jpg
      ├── 000043.txt
      ├── 000057.jpg
      ├── 000057.txt
      ├── 000070.jpg
      └── 000070.txt
  

• Generate a dataset path .txt file, the example content is as follows：

  train.txt

  /home/qiuqiu/Desktop/dataset/train/000001.jpg
  /home/qiuqiu/Desktop/dataset/train/000002.jpg
  /home/qiuqiu/Desktop/dataset/train/000003.jpg
  

  val.txt

  /home/qiuqiu/Desktop/dataset/val/000070.jpg
  /home/qiuqiu/Desktop/dataset/val/000043.jpg
  /home/qiuqiu/Desktop/dataset/val/000057.jpg
  

• Generate the .names category label file, the sample content is as follows:

  category.names

  person
  bicycle
  car
  motorbike
  ...
  
  

• The directory structure of the finally constructed training data set is as follows:

  .
  ├── category.names        # .names category label file
  ├── train                 # train dataset
  │   ├── 000001.jpg
  │   ├── 000001.txt
  │   ├── 000002.jpg
  │   ├── 000002.txt
  │   ├── 000003.jpg
  │   └── 000003.txt
  ├── train.txt              # train dataset path .txt file
  ├── val                    # val dataset
  │   ├── 000043.jpg
  │   ├── 000043.txt
  │   ├── 000057.jpg
  │   ├── 000057.txt
  │   ├── 000070.jpg
  │   └── 000070.txt
  └── val.txt                # val dataset path .txt file
  
  

Build the training .yaml configuration file

• Reference./configs/coco.yaml

  DATASET:
    TRAIN: "/home/qiuqiu/Desktop/coco2017/train2017.txt"  # Train dataset path .txt file
    VAL: "/home/qiuqiu/Desktop/coco2017/val2017.txt"      # Val dataset path .txt file 
    NAMES: "dataset/coco128/coco.names"                   # .names category label file
  MODEL:
    NC: 80                                                # Number of detection categories
    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                                          # ？？？？
    WARMUP: true                                          # Trun on warm up
    BATCH_SIZE: 64                                        # Batch size
    END_EPOCH: 350                                        # Train epichs
    MILESTIONES:                                          # Declining learning rate steps
      - 150
      - 250
      - 300
  

Train

• Perform training tasks

  python3 train.py --yaml configs/coco.yaml
  

Evaluation

• Calculate map evaluation

  python3 eval.py --yaml configs/coco.yaml --weight weights/weight_AP05:0.253207_280-epoch.pth
  

• COCO2017 evaluation

  creating index...
  index created!
  creating index...
  index created!
  Running per image evaluation...
  Evaluate annotation type *bbox*
  DONE (t=30.85s).
  Accumulating evaluation results...
  DONE (t=4.97s).
  Average Precision  (AP) @[ IoU=0.50:0.95 | area=   all | maxDets=100 ] = 0.130
  Average Precision  (AP) @[ IoU=0.50      | area=   all | maxDets=100 ] = 0.253
  Average Precision  (AP) @[ IoU=0.75      | area=   all | maxDets=100 ] = 0.119
  Average Precision  (AP) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.021
  Average Precision  (AP) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.129
  Average Precision  (AP) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.237
  Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=  1 ] = 0.142
  Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets= 10 ] = 0.208
  Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=100 ] = 0.214
  Average Recall     (AR) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.043
  Average Recall     (AR) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.236
  Average Recall     (AR) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.372
  
  

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_AP05:0.253207_280-epoch.pth --img data/3.jpg --onnx
  

Export torchscript

• You can export .pt by adding the --torchscript option when executing test.py

  python3 test.py --yaml configs/coco.yaml --weight weights/weight_AP05:0.253207_280-epoch.pth --img data/3.jpg --torchscript
  

NCNN

• Need to compile ncnn and opencv in advance and modify the path in build.sh

  cd example/ncnn/
  sh build.sh
  ./FastestDet
  

onnx-runtime

• You can learn about the pre and post-processing methods of FastestDet in this Sample

  cd example/onnx-runtime
  pip install onnx-runtime
  python3 runtime.py
  

Citation

• If you find this project useful in your research, please consider cite:

  @misc{=FastestDet,
        title={FastestDet: Ultra lightweight anchor-free real-time object detection algorithm.},
        author={xuehao.ma},
        howpublished = {\url{https://github.com/dog-qiuqiu/FastestDet}},
        year={2022}
  }
  

Reference

• https://github.com/Tencent/ncnn