Multi-object detection by lightweight CNN on embedded system.

Project abstract:

1. I mainly use those lightweight CNNs to detect hand pose on raspberry pi, but they can detect other objects easily too.
2. Some lightweight CNNs In the table below are reference to github project and I improved them, and others are reference to some papers.
3. I have created some scripts to generate voc formate dataset for caffe, it help me to extend dataset fast.

Lightweight CNNs	note	Supported System
MobileNet-SSD	Improved	Ubuntu and Raspbian
MobileNetV2-SSD	Haven't improved yet	Ubuntu
SqueezeNet-SSD	Significant improved	Ubuntu and Raspbian
ShuffleNet-SSD	Improved	Ubuntu with cuda
Tiny-SSD	Poor performance	Ubuntu and Raspbian

• Platform:

  • Training: Ubuntu
  • Application: Raspbian

• CNN Framwork: Caffe

• Dependencies: Opencv

• Reference CNN:

  • SSD:
    • Paper: https://arxiv.org/pdf/1512.02325.pdf
    • Github: https://github.com/weiliu89/caffe/tree/ssd
  • MobileNet:
    • Paper: https://arxiv.org/pdf/1704.04861.pdf
    • Github: https://github.com/tensorflow/models/tree/master/research/slim/nets
  • MobileNetV2:
    • Paper: https://128.84.21.199/pdf/1801.04381.pdf
    • Github: https://github.com/tensorflow/models/tree/master/research/slim/nets/mobilenet
    • Blog: https://ai.googleblog.com/2018/04/mobilenetv2-next-generation-of-on.html
  • SqueezeNet:
    • Paper: https://arxiv.org/pdf/1602.07360.pdf
    • Code: https://github.com/DeepScale/SqueezeNet
  • ShuffleNet:
    • Paper: https://arxiv.org/pdf/1707.01083.pdf
    • Code: https://github.com/farmingyard/ShuffleNet ; https://github.com/MG2033/ShuffleNet

• Reference github:

  • https://github.com/weiliu89/caffe
  • https://github.com/chuanqi305/MobileNet-SSD
  • https://github.com/chuanqi305/MobileNetv2-SSDLite
  • https://github.com/chuanqi305/SqueezeNet-SSD