Understanding deep networks by visualizing its evidence

Abstract
    이 연구에서는 각각 image classification과 action recognition에서 decision making을 할때 가장 activation이 활성화 되는 neuron들을 deconvolution을 통하여서 visualization하였습니다. 이를 통하여 decision making을 할 때 가장 중요한 역할을 한 특징들이 무엇인지 분석하고자 하였습니다.

Paper
    정승준. "Understanding deep networks by visualizing its evidence= 판단 근거 표면화를 통한 딥 네트워크의 이해." Master thesis (2016).

  • Visualizing evidences for a deep network's classification
    • CNN visualization

    • CNN+LSTM (LRCN) visualization
      • Original (ApplyEyeMakeup, UCF101 dataset)
        https://drive.google.com/open?id=1XhdVChGDdGo3jLKZlVaE8FW7Goa9zqHW


      • Visualized major evidences (8 frames per sequence for LRCN)
        https://drive.google.com/open?id=1vu_QhYdiACUDPzFSBDkjZCRxq9kgYFIj



      • Original ("time" in Korean Sign Language, self-generated dataset)
        https://drive.google.com/open?id=1xQgiOhEOeTCxjdqLdMFCchrT9YKbejOn

      • Visualized major evidences (8 frames per sequence for LRCN)
        https://drive.google.com/open?id=1-MxESewNiNTJ6gphZE131LsgIwXeAIEG