{
  "id": "1705.05922",
  "version": "v1",
  "published": "2017-05-16T21:05:49.000Z",
  "updated": "2017-05-16T21:05:49.000Z",
  "title": "LCDet: Low-Complexity Fully-Convolutional Neural Networks for Object Detection in Embedded Systems",
  "authors": [
    "Subarna Tripathi",
    "Gokce Dane",
    "Byeongkeun Kang",
    "Vasudev Bhaskaran",
    "Truong Nguyen"
  ],
  "comment": "Embedded Vision Workshop in CVPR",
  "categories": [
    "cs.CV"
  ],
  "abstract": "Deep convolutional Neural Networks (CNN) are the state-of-the-art performers for object detection task. It is well known that object detection requires more computation and memory than image classification. Thus the consolidation of a CNN-based object detection for an embedded system is more challenging. In this work, we propose LCDet, a fully-convolutional neural network for generic object detection that aims to work in embedded systems. We design and develop an end-to-end TensorFlow(TF)-based model. Additionally, we employ 8-bit quantization on the learned weights. We use face detection as a use case. Our TF-Slim based network can predict different faces of different shapes and sizes in a single forward pass. Our experimental results show that the proposed method achieves comparative accuracy comparing with state-of-the-art CNN-based face detection methods, while reducing the model size by 3x and memory-BW by ~4x comparing with one of the best real-time CNN-based object detector such as YOLO. TF 8-bit quantized model provides additional 4x memory reduction while keeping the accuracy as good as the floating point model. The proposed model thus becomes amenable for embedded implementations.",
  "revisions": [
    {
      "version": "v1",
      "updated": "2017-05-16T21:05:49.000Z"
    }
  ],
  "analyses": {
    "keywords": [
      "object detection",
      "low-complexity fully-convolutional neural networks",
      "embedded system",
      "cnn-based face detection methods",
      "real-time cnn-based object detector"
    ],
    "note": {
      "typesetting": "TeX",
      "pages": 0,
      "language": "en",
      "license": "arXiv",
      "status": "editable"
    }
  }
}