- Object Detection

Using automatic compression allows for easy model compression without a deep understanding of compression. Advanced compression offers higher flexibility in settings and may yield better compression results.

'Compressed' Type: Automatic Compression
'Compressed(Adv.)' Type: Advanced Compression

All of the original and compressed models can be downloaded easily on the Model Compressor Model Zoo.

PyTorch

{
  "data": {
    "h-0": "Model",
    "h-1": "Best Practice",
    "h-2": "Type",
    "h-3": "Dataset",
    "h-4": "mAP(0.5) (%)",
    "h-5": "mAP(0.5:0.95)(%)",
    "h-6": "FLOPs (M)",
    "h-7": "Params (M)",
    "h-8": "Latency (ms)",
    "h-9": "Model Size (MB)",
    "0-0": "[YOLOX](https://github.com/Megvii-BaseDetection/YOLOX)",
    "0-1": "",
    "0-2": "Original",
    "0-3": "COCO",
    "0-4": "68.0",
    "0-5": "49.7",
    "0-6": "156006.20",
    "0-7": "54.21",
    "0-8": "12239.46",
    "0-9": "207.37",
    "1-0": "YOLOX",
    "1-1": "[Google Colab](https://colab.research.google.com/github/Nota-NetsPresso/NetsPresso-CompressionToolkit-ModelZoo/blob/main/best_practices/object_detection/torch/yolox_coco/YOLOX.ipynb)",
    "1-2": "Compressed-1",
    "1-3": "COCO",
    "1-4": "67.16 (-0.84)",
    "1-5": "48.64 (-1.06)",
    "1-6": "101804.06 (1.53x)",
    "1-7": "19.96 (2.7x)",
    "1-8": "8502.72 (1.44x)",
    "1-9": "76.61 (2.7x)",
    "2-0": "YOLOX",
    "2-1": "[Google Colab](https://colab.research.google.com/github/Nota-NetsPresso/NetsPresso-CompressionToolkit-ModelZoo/blob/main/best_practices/object_detection/torch/yolox_coco/YOLOX.ipynb)",
    "2-2": "Compressed-2",
    "2-3": "COCO",
    "2-4": "61.43 (-6.57)",
    "2-5": "43.23 (-5.47)",
    "2-6": "38607.03 (4.04x)",
    "2-7": "4.93 (11.0x)",
    "2-8": "4235.37 (2.89x)",
    "2-9": "19.17 (10.80x)",
    "3-0": "YOLOv7",
    "3-1": "",
    "3-2": "Original",
    "3-3": "PASCAL VOC",
    "3-4": "89.6",
    "3-5": "71.3",
    "3-6": "104739.64",
    "3-7": "37.27",
    "3-8": "5464.59",
    "3-9": "146.33",
    "4-0": "YOLOv7",
    "4-1": "[Google Colab](https://colab.research.google.com/github/Nota-NetsPresso/NetsPresso-CompressionToolkit-ModelZoo/blob/main/best_practices/object_detection/torch/yolov7_voc/YOLOv7.ipynb)",
    "4-2": "Compressed-1",
    "4-3": "PASCAL VOC",
    "4-4": "88.4  \n(-1.2)",
    "4-5": "69.6  \n(-1.7)",
    "4-6": "77859.81  \n(1.35x)",
    "4-7": "17.12  \n(2.18x)",
    "4-8": "3855.95  \n(1.42x)",
    "4-9": "67.45  \n(2.00x)",
    "5-0": "YOLOv7",
    "5-1": "[Google Colab](https://colab.research.google.com/github/Nota-NetsPresso/NetsPresso-CompressionToolkit-ModelZoo/blob/main/best_practices/object_detection/torch/yolov7_voc/YOLOv7.ipynb)",
    "5-2": "Compressed-2",
    "5-3": "PASCAL VOC",
    "5-4": "85.2  \n(-4.4)",
    "5-5": "63.6  \n(-7.7)",
    "5-6": "21878.87  \n(4.79x)",
    "5-7": "2.55  \n(14.60x)",
    "5-8": "2041.65  \n(2.68x)",
    "5-9": "10.61  \n(13.79x)"
  },
  "cols": 10,
  "rows": 6,
  "align": [
    "left",
    "left",
    "left",
    "left",
    "left",
    "left",
    "left",
    "left",
    "left",
    "left"
  ]
}

The model’s latency is measured on Raspberry Pi 4B (1.5GHz ARM Cortex).
Options: FP32, ONNX runtime

TensorFlow-Keras

Model	Best Practice	Type	Dataset	mAP(0.5) (%)	mAP(0.5:0.95)(%)	FLOPs (M)	Params (M)	Latency (ms)	Model Size (MB)
YOLOv4		Original	PASCAL VOC	82.22	-	61871.82	65.32	64318.70	262.90
YOLOv4	Google Colab	Compressed-1	PASCAL VOC	87.23 (+5.01)	-	11459.69 (5.4x)	10.59 (6.17x)	28651.70 (2.16x)	44.12 (5.96x)
YOLOv4	Google Colab	Compressed-2	PASCAL VOC	87.91 (+5.69)	-	14442.96 (4.28x)	10.71 (6.1x)	28976.40 (2.14x)	44.36 (5.93x)

YOLOv4 model with EfficientNet B1 based backbone.
The model’s latency is measured on Raspberry Pi 4B (1.5GHz ARM Cortex).
Options: FP32, TFLite